# Estimates on training vs. validation samples

Before moving to cross-validation, it was natural to say “I will burn 50% (say) of my data to train a model, and then use the remaining to fit the model”. For instance, we can use training data for variable selection (e.g. using some stepwise procedure in a logistic regression), and then, once variable have been selected, fit the model on the remaining set of observations. A natural question is usually “does it really matter ?”.

In order to visualize this problem, consider my (simple) dataset

MYOCARDE=read.table( "http://freakonometrics.free.fr/saporta.csv", head=TRUE,sep=";")

Let us generate 100 training samples (where we keep about 50% of the observations). On each of them, we use a stepwise procedure, and we keep the estimates of the remaining variables (and their standard deviation actually)

n=nrow(MYOCARDE) M=matrix(NA,100,ncol(MYOCARDE)) colnames(M)=c("(Intercept)",names(MYOCARDE)[1:7]) S1=S2=M1=M2=M for(i in 1:100){ idx = which(sample(0:1,size=n, replace=TRUE)==1) reg=step(glm(PRONO=="DECES"~.,data=MYOCARDE[idx,])) nm=names(reg$coefficients) M1[i,nm]=reg$coefficients S1[i,nm]=summary(reg)$coefficients[,2] f=paste("PRONO=='DECES'~",paste(nm[-1],collapse="+"),sep="") reg=glm(f,data=MYOCARDE[-idx,]) M2[i,nm]=reg$coefficients S2[i,nm]=summary(reg)$coefficients[,2] } Then, for the 7 covariates (and the constant) we can look at the value of the coefficient in the model fitted on the training sample, and the value on the model fitted on the validation sample (of course, only when they were remaining) for(j in 1:8){ idx=which(!is.na(M1[,j])) plot(M1[idx,j],M2[idx,j]) abline(a=0,b=1,lty=2,col="gray") segments(M1[idx,j]-2*S1[idx,j],M2[idx,j],M1[idx,j]+2*S1[idx,j],M2[idx,j]) segments(M1[idx,j],M2[idx,j]-2*S2[idx,j],M1[idx,j],M2[idx,j]+2*S2[idx,j]) } For instance, with the intercept, we have the following where horizontal segments are confidence intervals of the parameter on the model fitted on the training sample, the vertical on the validation sample. The green part means some sort of consistency, while the red one means that actually, the coefficient was negative with one model, positive with the other one. Which is odd (but in that case, observe that coefficients are rarely significant). We can also visualize the joint distribution of the two estimators, for(j in 1:8){ library(ks) idx = which(!is.na(M1[,j])) Z = cbind(M1[idx,j],M2[idx,j]) H = Hpi(x=Z) fhat = kde(x=Z, H=H) image(fhat$eval.points[[1]], fhat$eval.points[[2]],fhat$estimate) abline(a=0,b=1,lty=2,col="gray") abline(v=0,lty=2) abline(h=0,lty=2) }

which are here, almost on the diagonal,

meaning that the intercept on the two samples is (more or less) the same. We can then look at other parameters (which is actually more interesting).

On that variable, it seems that it is significant on the training dataset (somehow, it is consistent with the fact that it is remaining in the model after the stepwise procedure) but not on the validation sample (or hardly significant).

Others are much more consistent (with some possible outliers)

On the next one, we have again significance on the training sample, but not on the validation sample,

and probably more interesting

where the two are very consistent.

# Variance of the slope in a regression model

In my “applied linear models” exam, there was a tricky question (it was a multiple choice, so no details were asked). I was simply asking if the following statement was valid, or not

Consider a linear regression with one single covariate, $y=\beta_0+\beta_1x_1+\varepsilon$ and the least-square estimates. The variance of the slope is $\text{Var}[\widehat{\beta}_1]$ Do we decrease this variance if we add one variable, and consider $y=\beta_0+\beta_1x_1+\beta_2x_2+\varepsilon$ ?

For the exam, the expected answer was simply “no”. In a nutshell, there are two cases where we should expect different changes,

• if $x_1$ and $x_2$ are highly correlated, then we should expect the variance to increase
• if $x_1$ and $x_2$ are not correlated, then we should expect the variance to decrease

We did briefly observed (and discussed) those points on examples during the lecture… but I wanted to go a bit further, since I couldn’t find any analytical results. Let us generate a model $y=\beta_0+\beta_1x_1+\beta_2x_2+\varepsilon$, and then compare the variance $\text{Var}[\widehat{\beta}_1]$ on the two fitted modes, depending on the correlation between $x_1$ and $x_2$

library(mnormt) n=200 s=function(r=0){ S=matrix(c(1,r,r,1),2,2) X=rmnorm(n,c(0,0),S) B=data.frame(y=-2+X[,1]+X[,2]+rnorm(n)/2, x1=X[,1], x2=X[,2]) reg12=lm(y~x1+x2,data=B) reg1=lm(y~x1,data=B) k=summary(reg12)$coefficients[2,2]/summary(reg1)$coefficients[2,2] k}

Let us generate 500 samples for each value of the correlation, from -0.9 to _0.9

M=NULL for(r in ((-9):9)/10) M=cbind(M,Vectorize(s)(rep(r,500)))

and let us plot the ratio of the two variances

plot(0:1,0:1,xlim=c(-1,1),ylim=c(0,2),col="white") for(i in 1:19) points(rep((((-9):9)/10)[i],500),M[,i],col="light blue") VM=apply(M,2,mean) lines((((-9):9)/10),VM,col="red",lwd=2) abline(h=1,lty=2)

If the ratio exceeds 1, the variance increases when adding a covariate.

Indeed, here, when the two variables are independent, the variance is divided by two. But when covariates are highly correlated, the variance is multiplied by two…

Now, what if, actually, $x_2$ is not a real explanatory variable : the true model we generate is $y=\beta_0+\beta_1x_1+\varepsilon$. In that case,

s=function(r=0){ S=matrix(c(1,r,r,1),2,2) X=rmnorm(n,c(0,0),S) B=data.frame(y=-2+X[,1]+rnorm(n)/2, x1=X[,1], x2=X[,2]) reg12=lm(y~x1+x2,data=B) reg1=lm(y~x1,data=B) k=summary(reg12)$coefficients[2,2]/summary(reg1)$coefficients[2,2] k}

we get our samples as previously

M=NULL for(r in ((-9):9)/10) M=cbind(M,Vectorize(s)(rep(r,500)))

and we plot those ratios

plot(0:1,0:1,xlim=c(-1,1),ylim=c(0,2),col="white") for(i in 1:19) points(rep((((-9):9)/10)[i],500),M[,i],col="light blue") VM=apply(M,2,mean) lines((((-9):9)/10),VM,col="red",lwd=2) abline(h=1,lty=2)

In the case we add a useless variable $x_2$, whatever the correlation with $x_1$, it will always, on average, increase the variance of $\widehat{\beta}_1$.

# Random thoughts on econometric models with (pure) random features

For my lectures on applied linear models, I wanted to illustrate the fact that the $R^2$ is never a good measure of the goodness of the model, since it’s quite easy to improve it. Consider the following dataset

n=100 df=data.frame(matrix(rnorm(n*n),n,n)) names(df)=c("Y",paste("X",1:99,sep=""))

with one variable of interest $y$, and 99 features $x_j$. All of them being (by construction) independent. And we have 100 observations… Consider here the regression on the first $k$ features, and compute $R_k^2$ of that regression

reg=function(k){ frm=paste("Y~",paste("X",1:k,collapse="+",sep="")) model=lm(frm,data=df) summary(model)$adj.r.squared} Let us see what’s going on… plot(1:99,Vectorize(reg)(1:99)) (actually, it’s not exactly what we have on the graph…. we have the average obtained over 1,000 samples randomly generated, with 90% confidence bands). Oberve that $\mathbb{E}[R^2_k]=k/n$, i.e. if we add some pure random noise, we keep increasing the $R^2$ (up to 1, actually). Good news, as we’ve seen in the course, the adjusted $R^2$ – denoted $\bar R^2$-might help. Observe that $\mathbb{E}[\barR^2_k]=0$, so, in some sense, adding features does not help here… reg=function(k){ frm=paste("Y~",paste("X",1:k,collapse="+",sep="")) model=lm(frm,data=df) summary(model)$r.squared} plot(1:99,Vectorize(reg)(1:99))

We can actually do the same with Akaike criteria $AIC_k$ and Schwarz (bayesian) criteria $BIC_k$.

reg=function(k){ frm=paste("Y~",paste("X",1:k,collapse="+",sep="")) model=lm(frm,data=df) AIC(model)} plot(1:99,Vectorize(reg)(1:99))

For the $AIC$, the intitial increase makes sense : we should not prefer the model with 10 covariates, compared with nothing. The strange thing is the far right behavior : we prefer here 80 random noise features to none ! Which I find hard to interprete… For the $BIC$ the code is simply

reg=function(k){ frm=paste("Y~",paste("X",1:k,collapse="+",sep="")) model=lm(frm,data=df) BIC(model)} plot(1:99,Vectorize(reg)(1:99))

and here also, we have the same pattern, where we prefer a big model with juste pure noise to nothing…

A last one to conclude (or not) : what about the leave-one-out cross validation mean squared error ? More precisely, $$CV=\frac{1}{n}\sum_{i=1}\widehat{\varepsilon}^2_{-i}$$where $\widehat{\varepsilon}^2_{-i}=y_i-\widehat{y}_{-i}$ where $\widehat{y}_{-i}$ is the predicted value obtained with the model is estimated when the $i$th observation is deleted. One can prove that $$\widehat{\beta}_{-i}=\widehat{\beta}-(\mathbf{X}^T\mathbf{X})^{-1}\mathbf{x}_i\hat\varepsilon_i(1-H_{i,i})^{-1}$$where $H$ is the classical hat matrix, thus$$\widehat{\varepsilon}_{-i}=(1-H_{i,i})^{-1}\hat\varepsilon_i$$i.e. we do note have to estimate (at each round) $n$ models

reg=function(k){ frm=paste("Y~",paste("X",1:k,collapse="+",sep="")) model=lm(frm,data=df) h=lm.influence(model)$hat/2 mean( (residuals(model)/1-h)^2 ))} plot(1:99,Vectorize(reg)(1:99)) Here, it make sense : adding noisy features yields overfit ! So the mean squared error is decreasing ! That’s all nice, but it might not be very realistic… Here, for my model with only one variable, I just pick one, at random…. In practice, we try to get the “best one”… So a more natural idea would be to order the variables according to their correlations with $y$, df=data.frame(matrix(rnorm(n*n),n,n)) df=df[,rev(order(abs(cor(df)[1,])))] names(df)=c("Y",paste("X",1:99,sep=""))} and as before, we can plot the evolution of $R^2_k$ as a function of $k$ the number of features considered, which is increasing, with a higher slope at the beginning… For the $\bar R^2_k$ we might actually prefer a correlated noise to nothing (which makes sense actually). So here since we somehow chose our variables, $\bar R^2_k$ seems to be always positive… For the $AIC_k$ here also, there is an improvement. Before coming back to the original situation (with about 80 features) and here also, we observe the drop on the far right part of the graph The $BIC_k$ might like the top three features, but soon, we have a deterioration…. even if here also, we have the drop at the far right (with more than 95 features… for 100 observations). Finally, observe that here again, our (leave-one-out) cross-validation has not been mesled by our noisy variables : it is always decreasing ! So it seems that cross-validation techniques are more robust than the $AIC$ and $BIC$ (even if we mentioned in a previous post connexions between all those concepts) when we have a lot a noisy (non-relevent) features. # Foundations of Machine Learning, part 5 This post is the nineth (and probably last) one of our series on the history and foundations of econometric and machine learning models. The first fours were on econometrics techniques. Part 8 is online here. ## Optimization and algorithmic aspects In econometrics, (numerical) optimization became omnipresent as soon as we left the Gaussian model. We briefly mentioned it in the section on the exponential family, and the use of the Fisher score (gradient descent) to solve the first order condition $\mathbf{X}^T W(\beta)^{-1})[y-\widehat{y}]=\mathbf{0}$. In learning, optimization is the central tool. And it is necessary to have effective optimization algorithms, to solve problems (described previously) of the form: $$\widehat{\beta}\in\underset{\beta\in\mathbb{R}^p}{\text{argmin}}\left\lbrace\sum_{i=1}^n \ell(y_i,\beta_0+\mathbf{x}^T\beta)+\lambda\Vert\boldsymbol{\beta}\Vert\right\rbrace$$In some cases, instead of global optimization, it is sufficient to consider optimization by coordinates (widely studied in Daubechies et al. (2004)). If $f:\mathbb{R}^d\rightarrow\mathbf{R}$ is convex and differentiable, if $\mathbf{x}$ satisfies $f(\mathbf{x}+h\boldsymbol{e}_i)\geq f(\mathbf{x})$ for any $h>0$ and $i\in\{1,\cdots, d\}$then $f(\mathbf{x})=\min\{f\}$, where $\mathbf{e}=(\mathbf{e}_i)$ is the canonical basis of $\mathbb{R}^d$. However, this property is not true in the non-differentiable case. But if we assume that the non-differentiable part is separable (additively), it becomes true again. More specifically, if$$f(\mathbf{x})=g(\mathbf{x})+\sum_{i=1}^d h_i(x_i)$$with$$\left\lbrace\begin{array}{l}g: \mathbb{R}^d\rightarrow\mathbb{R}\text{ convex-differentiable}\\h_i: \mathbb{R}\rightarrow\mathbb{R}\text{ convex}\end{array}\right.$$This was the case for Lasso regression, $\beta)\mapsto\| \mathbf{y}-\beta_0-\mathbf{X}\beta\|_{\ell_2 }+\lambda\|\beta\|_{\ell_1}$, as shown by Tsen (2001). Getting back to our initial notations, we can use a coordinate descent algorithm: from an initial value $\mathbf{x}^{(0)}$, we consider (by iterating)$$x_j^{(k)}\in\text{argmin}\big\lbrace f(x_1^{(k)},\cdots,x_{k-1}^{(k)},x_k,x_{k+1}^{(k-1)},\cdots,x_n^{(k-1)})\big\rbrace$$ for $j=1,2,\cdots,n$These algorithmic problems and numerical issues may seem secondary to econometricians. However, they are essential in automatic learning: a technique is interesting if there is a stable and fast algorithm, which allows to obtain a solution. These optimization techniques can be transposed: for example, this coordinate descent technique can be used in the case of SVM methods (known as “vector support” methods) when the space is not linearly separable, and the classification error must be penalized (we will come back to this technique in the next section). ## In-sample, out-of-sample and cross-validation These techniques seem intellectually interesting, but we have not yet discussed the choice of the penalty parameter $\lambda$. But this problem is actually more general, because comparing two parameters $\widehat{\beta}_{\lambda_1}$ and $\widehat{\beta}_{\lambda_2}$ is actually comparing two models. In particular, if we use a Lasso method, with different thresholds $\lambda$, we compare models that do not have the same dimension. Previously, we have addressed the problem of model comparison from an econometric perspective (by penalizing overly complex models). In the learning literature, judging the quality of a model on the data used to construct it does not make it possible to know how the model will behave on new data. This is the so-called “generalization” problem. The traditional approach then consists in separating the sample (size $n$) into two parts: a part that will be used to train the model (the training database, in-sample, size $m$) and a part that will be used to test the model (the testing database, out-of-sample, size $n-m$). The latter then makes it possible to measure a real predictive risk. Suppose that the data are generated by a linear model $y_i=\mathbf{x}_i^T \beta_0+\varepsilon_i$ where $\varepsilon_i$ are independent and centred law achievements. The empirical quadratic risk in-sample is here$$\frac{1}{m}\sum_{i=1}^m\mathbb{E}\big([\mathbf{x}_i^T \widehat{\beta}-\mathbf{x}_i^T \beta_0]^2\big)=\mathbb{E}\big([\mathbf{x}_i^T \widehat{\beta}-\mathbf{x}_i^T \beta_0]^2\big),$$for any observation $i$. Assuming the residuals $\varepsilon$ Gaussian, then we can show that this risk is worth $\sigma^2 \text{trace} (\Pi_X)/m$ is $\sigma^2 p/m$. On the other hand, the empirical out-of-sample quadratic risk is here $$\mathbb{E}\big([\mathbf{x}^T \widehat{\beta}-\mathbf{x}^T \beta_0]^2\big)$$where $\mathbf{x}$ is a new observation, independent of the others. It can be noted that $$\mathbb{E}\big([\mathbf{x}^T \widehat{\beta}-\mathbf{x}^T \beta_0]^2\big\vert \mathbf{x}\big)=\text{Var}\big(\mathbf{x}^T \widehat{\beta}\big\vert \mathbf{x}\big)=\sigma^2\mathbf{x}^T(\mathbf{x}^T\mathbf{x})^{-1}\mathbf{x},$$and by integrating with respect to $\mathbf{x}$, $$\mathbb{E}\big([\mathbf{x}^T \widehat{\beta}-\mathbf{x}^T\beta_0]^2\big)=\sigma^2\text{trace}\big(\mathbb{E}[\mathbf{x}\mathbf{x}^T]\mathbb{E}\big[(\mathbf{x}^T\mathbf{x})^{-1}\big]\big).$$The expression is then different from that obtained in-sample, and using the Groves & Rothenberg (1969) increase, we can show that $$\mathbb{E}\big([\mathbf{x}^T \widehat{\beta}-\mathbf{x}^T \beta_0]^2\big) \geq \sigma^2\frac{p}{m}$$which is pretty intuitive, when we start thinking about it. Except in some simple cases, there is no simple (explicit) formula. Note, however, that if $\mathbf{X}\sim\mathcal{N}(0,\sigma^2 \mathbb{I})$, then $\mathbf{x}^T \mathbf{x}$ follows a Wishart law, and it can be shown that $$\mathbb{E}\big([\mathbf{x}^T \widehat{\beta}-\mathbf{x}^T \beta_0]^2\big)=\sigma^2\frac{p}{m-p-1}.$$If we now look at the empirical version: if $\widehat{\beta}$ is estimated on the first $m$ observations,$$\widehat{\mathcal{R}}^{~\text{ IS}}=\sum_{i=1}^m [y_i-\boldsymbol{x}_i^T\widehat{\boldsymbol{\beta}}]^2\text{ and }\widehat{\mathcal{R}}^{\text{ OS}}=\sum_{i=m+1}^{n} [y_i-\boldsymbol{x}_i^T\widehat{\boldsymbol{\beta}}]^2$$and as Leeb (2008) noted, $\widehat{\mathcal{R}}^{\text{IS}}-\widehat{\mathcal{R}}^{\text{OS}}\approx 2\cdot\nu$ where $\nu$ represents the number of degrees of freedom, which is not unlike the penalty used in the Akaike test. Figure 4 shows the respective evolution of $\widehat{\mathcal{R}}^{\text{IS}}$ and $\widehat{\mathcal{R}}^{\text{OS}}$ according to the complexity of the model (number of degrees in a polynomial regression, number of nodes in splines, etc). The more complex the model, the more $\widehat{\mathcal{R}}^{\text{IS}}$ will decrease (this is the red curve, below). But that’s not what we’re interested in here: we want a model that predicts well on new data (i. e. out-of-sample). As Figure 4 shows, if the model is too simple, it does not predict well (as it does with in-sample data). But what we can see is that if the model is too complex, we are in a situation of “overlearning”: the model will start to model the noise. Of course, this figure should remind us of the one we’ve seen in our second post of that series Figure 4 : Generalization, under- and over-fitting Instead of splitting the database in two, with some of the data that will be used to calibrate the model and some to study its performance, it is also possible to use cross-validation. To present the general idea, we can go back to the “jackknife”, introduced by Quenouille (1949) (and formalized by Quenouille (1956) and Tukey (1958)) relatively used in statistics to reduce bias. Indeed, if we assume that $\{y_1,\cdots,y_n\}$ is a sample drawn according to a law $F_\theta$, and that we have an estimator $T_n (\mathbf{y})=T_n (y_1,\cdots,y_n)$, but that this estimator is biased, with $\mathbf{E}[T_n (\mathbf{Y})]=\theta+O(n^{-1})$, it is possible to reduce the bias by considering $$\widetilde{T}_n(\mathbf{y})=\frac{1}{n}\sum_{i=1}^n T_{n-1}(\mathbf{y}_{(i)})\text{ where }\mathbf{y}_{(i)}=(y_1,\cdots,y_{i-1},y_{i+1},\cdots,y_n)$$It can then be shown that $\mathbb{E}[\tilde{T}_n(Y)]=\theta+O(n^{-2})$The idea of cross-validation is based on the idea of building an estimator by removing an observation. Since we want to build a predictive model, we will compare the forecast obtained with the estimated model, and the missing observation$$\widehat{\mathcal{R}}^{\text{ CV}}=\frac{1}{n}\sum_{i=1}^n \ell(y_i,\widehat{m}_{(i)}(\mathbf{x}_i))$$We will speak here of the “leave-one-out” (loocv) method. This technique reminds us of the traditional method used to find the optimal parameter in exponential smoothing methods for time series. In simple smoothing, we will construct a forecast from a time series as ${}_t\widehat{y}_{t+1} =\alpha\cdot{}_{t-1}\widehat{y}_t +(1-\alpha)\cdot y_t$, where $\alpha\in[0,1]$, and we will consider as “optimal” $$\alpha^\star = \underset{\alpha\in[0,1]}{\text{argmin}}\left\lbrace \sum_{t=2}^T \ell({}_{t-1}\widehat{y}_{t},y_{t}) \right\rbrace$$as described by Hyndman et al (2009). The main problem with the leave-one-out method is that it requires calibration of n models, which can be problematic in large dimensions. An alternative method is cross validation by $k$-blocks (called “$k$-fold cross validation”) which consists in using a partition of $\{1,\cdots,n\}$ in $k$ groups (or blocks) of the same size, $\mathcal{I}_1,\cdots,\mathcal{I}_k$, and let us note $\mathcal{I}_{\bar j}=\{1,\cdots,n\}\setminus \mathcal{I}_j$. By noting $\widehat{m}_{(j)}$ built on the sample $\mathcal{I}_{\bar j}$, we then set:$$\widehat{\mathcal{R}}^{k-\text{ CV}}=\frac{1}{k}\sum_{j=1}^k \mathcal{R}_j\text{ where }\mathcal{R}_j=\frac{k}{n}\sum_{i\in\mathcal{I}_{{j}}} \ell(y_i,\widehat{m}_{(j)}(\mathbf{x}_i))$$Standard cross-validation, where only one observation is removed each time (loocv), is a special case, with $k=n$. Using $k=5$ or $10$ has a double advantage over $k=n$: (1) the number of estimates to be made is much smaller, 5 or 10 rather than $n$; (2) the samples used for estimation are less similar and therefore less correlated to each other, which tends to avoid excess variance, as recalled by James et al. (2013). Another alternative is to use boosted samples. Let $\mathcal{I}_b$ be a sample of size $n$ obtained by drawing with replacement in $\{1,\cdots,n\}$ to know which observations $(y_i,\mathbf{x}_i)$ will be kept in the learning population (at each draw). Note $\mathcal{I}_{\bar b}=\{1,\cdots,n\}\setminus\mathcal{I}_b$. By noting $\widehat{m}_{(b)}$ built on sample $\mathcal{I}_b$, we then set :$$\widehat{\mathcal{R}}^{\text{ B}}=\frac{1}{B}\sum_{b=1}^B \mathcal{R}_b\text{ where }\mathcal{R}_b=\frac{n_{\overline{b}}}{n}\sum_{i\in\mathcal{I}_{\overline{b}}} \ell(y_i,\widehat{m}_{(b)}(\mathbf{x}_i))$$where $n_{\bar b}$ is the number of observations that have not been kept in $\mathcal{I}_b$. It should be noted that with this technique, on average $e^{-1}\sim36.7\%$ of the observations do not appear in the boosted sample, and we find an order of magnitude of the proportions used when creating a calibration sample, and a test sample. In fact, as Stone (1977) had shown, the minimization of AIC is to be compared to the cross-validation criterion, and Shao (1997) showed that the minimization of BIC corresponds to $k$-fold cross-validation, with $k=n/\log n$. All those techniques here are mentioned in the “machine learning” section since they rely on automatic, computational techniques, and no probabilistic foundations are necessary. In many cases we did use the notation $m^\star$ (at least in the first posts on “machine learning” techniques) to highlight the fact that we want some sort of “optimal” model – and to make a distinction with estimators $\widehat{m}$ considered earlier, when we had some probabilistic framework. But of course, it is possible (and necessary) to build bridges between those two cultures… References are online here. As explained in the introduction, it is some sort of online version of an introduction to our joint paper with Emmanuel Flachaire and Antoine Ly, Econometrics and Machine Learning (initially writen in French), that will actually appear soon in the journal Economics and Statistics (in English and in French). # Foundations of Machine Learning, part 1 This post is the fifth one of our series on the history and foundations of econometric and machine learning models. The first fours were on econometrics techniques. Part 4 is online here. In parallel with these tools developed by, and for economists, a whole literature has been developed on similar issues, centered on the problems of prediction and forecasting. For Breiman (2001a), a first difference comes from the fact that the statistic has developed around the principle of inference (or to explain the relationship linking $y$ to variables $\mathbf{x}$) while another culture is primarily interested in prediction. In a discussion that follows the article, David Cox states very clearly that in statistic (and econometrics) “predictive success (…) is not the primary basis for model choice“. We will get back here on the roots of automatic learning techniques. The important point, as we will see, is that the main concern of machine learning is related to the generalization properties of a model, i.e. its performance – according to a criterion chosen a priori – on new data, and therefore on non-sample tests. ## A learning machine Today, we speak of “machine learning” to describe a whole set of techniques, often computational, as alternatives to the classical econometric approach. Before characterizing them as much as possible, it should be noted that historically other names have been given. For example, Friedman (1997) proposes to make the link between statistics (which closely resemble econometric techniques – hypothesis testing, ANOVA, linear regression, logistics, GLM, etc.) and what was then called “data mining” (which then included decision trees, methods from the closest neighbours, neural networks, etc.). The bridge between those two cultures corresponds to “statistical learning” techniques described in Hastie et al (2009). But one should keep in mind that machine learning is a very large field of research. The so-called “natural” learning (as opposed to machine learning) is that of children, who learn to speak, read and play. Learning to speak means segmenting and categorizing sounds, and associating them with meanings. A child also learns simultaneously the structure of his or her mother tongue and acquires a set of words describing the world around him or her. Several techniques are possible, ranging from rote learning, generalization, discovery, more or less supervised or autonomous learning, etc. The idea in artificial intelligence is to take inspiration from the functioning of the brain to learn, to allow “artificial” or “automatic” learning, by a machine. A first application was to teach a machine to play a game (tic-tac-toe, chess, go, etc.). An essential step is to explain the objective it must achieve to win. One historical approach has been to teach the machine the rules of the game. If it allows you to play, it will not help the machine to play well. Assuming that the machine knows the rules of the game, and that it has a choice between several dozen possible moves, which one should it choose? The classical approach in artificial intelligence uses the so-called min-max algorithm using an evaluation function: in this algorithm, the machine searches forward in the possible moves tree, as far as the calculation resources allow (about ten moves in chess, for example). Then, it calculates different criteria (which have been previously indicated to her) for all positions (number of pieces taken, or lost, occupancy of the center, etc. in our example of the chess game), and finally, the machine plays the move that allows it to maximize its gain. Another example may be the classification and recognition of images or shapes. For example, the machine must identify a number in a handwritten handwriting (checks, ZIP code on envelopes, etc). It is a question of predicting the value of a variable $y$, knowing that a priori $y\in\{0,1,2,\cdots,8,9\}$. A classical strategy is to provide the machine with learning bases, in other words here millions of labelled (identified) images of handwritten numbers. A simple (and natural) strategy is to use a decision criterion based on the closest neighbors whose labels are known (using a predefined metric). The method of the closest neighbors (“$k$-nearest neighbors”) can be described as follows: we consider (as in the previous part) a set of n observations, i. e. pairs $(y_i,\mathbf{x}_i)$ with $\mathbf{x}_i\in\mathbb{R}^p$. Let us consider a distance $\Delta$ on $\mathbb{R}^p$ (the Euclidean distance or the Mahalanobis distance, for example). Given a new observation $\mathbf{x}\in\mathbb{R}^p$, let us assume the ordered observations as a function of the distance between the $\mathbf{x}_i$ and $\mathbf{x}$, in the sense that $$\Delta(\mathbf{x}_1, \mathbf{x})\leq\Delta(\mathbf{x}_2, \mathbf{x})\leq\cdots\leq\Delta(\mathbf{x}_n, \mathbf{x})$$ then we can consider as prediction for y the average of the nearest $k$ neighbours,$$\widehat{m}_k(\mathbf{x})=\frac{1}{k}\sum_{i=1}^k y_i$$Learning here works by induction, based on a sample (called the learning – or training – sample). Automatic learning includes those algorithms that give computers the ability to learn without being explicitly programmed (as Arthur Samuel defined it in 1959). The machine will then explore the data with a specific objective (such as searching for the nearest neighbours in the example just described). Tom Mitchell proposed a more precise definition in 1998: a computer program is said to learn from experience $E$ in relation to a task $T$ and a performance measure $P$, if its performance on $T$, measured by $P$, improves with experience $E$. Task $T$ can be a defect score for example, and performance $P$ can be the percentage of errors made. The system learns if the percentage of predicted defects increases with experience. As we can see, machine learning is basically a problem of optimizing a criterion based on data (from now on called learning). Many textbooks on machine learning techniques propose algorithms, without ever mentioning any probabilistic model. In Watt et al (2016) for example, the word “probability” is mentioned only once, with this footnote that will surprise and make smile any econometricians, “the logistic regression can also be interpreted from a probabilistic perspective” (page 86). But many recent books offer a review of machine learning approaches using probabilistic theories, following the work of Vaillant and Vapnik. By proposing the paradigm of “probably almost correct” learning (PAC), a probabilistic flavor has been added to the previously very computational approach, by quantifying the error of the learning algorithm (usually in a classification problem). To be continued (references are online here)… # Probabilistic Foundations of Econometrics, part 4 This post is the fourth one of our series on the history and foundations of econometric and machine learning models. Part 3 is online here. ## Goodness of Fit, and Model In the Gaussian linear model, the determination coefficient – noted $R^2$ – is often used as a measure of fit quality. It is based on the variance decomposition formula $$\underbrace{\frac{1}{n}\sum_{i=1}^n (y_i-\bar{y})^2}_{\text{total variance}}=\underbrace{\frac{1}{n}\sum_{i=1}^n (y_i-\widehat{y}_i)^2}_{\text{residual variance}}+\underbrace{\frac{1}{n}\sum_{i=1}^n (\widehat{y}_i-\bar{y})^2}_{\text{explained variance}}$$ The $R^2$ is defined as the ratio of explained variance and total variance, another interpretation of the coefficient that we had introduced from the geometry of the least squares $$R^2= \frac{\sum_{i=1}^n (y_i-\bar{y})^2-\sum_{i=1}^n (y_i-\widehat{y}_i)^2}{\sum_{i=1}^n (y_i-\bar{y})^2}$$The sums of the error squares in this writing can be rewritten as a log-likelihood. However, it should be remembered that, up to one additive constant (obtained with a saturated model) in generalized linear models, deviance is defined by ${Deviance}(\widehat{\beta}) = -2\log[\mathcal{L}]$ which can also be noted $Deviance(\widehat{\mathbf{y}})$. A null deviance can be defined as the one obtained without using the explanatory variables $\mathbf{x}$, so that $\widehat{y}_i=\overline{y}$. It is then possible to define, in a more general context (with a non-Gaussian distribution for $y$)$$R^2=\frac{{Deviance}(\overline{y})-{Deviance}(\widehat{\mathbf{y}})}{{Deviance}(\overline{y})}=1-\frac{{Deviance}(\widehat{\mathbf{y}})}{{Deviance}(\overline{y})}$$However, this measure cannot be used to choose a model, if one wishes to have a relatively simple model in the end, because it increases artificially with the addition of explanatory variables without significant effect. We will then tend to prefer the adjusted $R^2$,$$\bar R^2 = {1-(1-R^{2})\cdot{n-1 \over n-p}} = R^{2}-\underbrace{(1-R^{2})\cdot{p-1 \over n-p}}_{\text{penalty}}$$where $p$ is the number of parameters of the model. Measuring the quality of fit will penalize overly complex models. This idea will be found in the Akaike criterion, where $AIC=Deviance+2\cdot p$ or in the Schwarz criterion, $BIC=Deviance+log(n)\cdot p$. In large dimensions (typically $p>\sqrt{n}$), we will tend to use a corrected $AIC$, defined by $AIC_c=Deviance+2⋅p⋅n/(n-p-1)$. These criterias are used in so-called “stepwise” methods, introducing the set methods. In the “forward” method, we start by regressing to the constant, then we add one variable at a time, retaining the one that lowers the $AIC$ criterion the most, until adding a variable increases the AIC criterion of the model. In the “backward” method, we start by regressing on all variables, then we remove one variable at a time, removing the one that lowers the $AIC$ criterion the most, until removing a variable increases the $AIC$ criterion from the model. Another justification for this notion of penalty (we will come back to this idea in machine learning) can be the following. Let us consider an estimator in the class of linear predictors,$$\mathcal{M}=\big\lbrace m:~m(\mathbf{x})=s_h(\mathbf{x})^T\mathbf{y} \text{ where }S=(s(\mathbf{x}_1),\cdots,s(\mathbf{x}_n))^T\text{ is some smoothing matrix}\big\rbrace$$ and assume that $y=m_0 (x)+\varepsilon$, with $\mathbb{E}[\varepsilon]=0$ and $Var[\varepsilon]=\sigma^2\mathbb{I}$, so that $m_0 (x)=\mathbb{E}[Y|X=x]$. From a theoretical point of view, the quadratic risk, associated with an estimated model $\widehat{m}$, $\mathbb{E}\big[(Y-\widehat{m}(\mathbf{X}))^2\big]$, is written$$\mathcal{R}(\widehat{m})=\underbrace{\mathbb{E}\big[(Y-m_0(\mathbf{X}))^2\big]}_{\text{error}}+\underbrace{\mathbb{E}\big[(m_0(\mathbf {X})-\mathbb{E}[\widehat{m}(\mathbf{X})])^2\big]}_{\text{bias}^2}+\underbrace{\mathbb{E}\big[(\mathbb{E}[\widehat{m}(\mathbf{X})]-\widehat{m}(\mathbf{X}))^2\big]}_{\text{variance}}$$if $m_0$ is the true model. The first term is sometimes called “Bayes error”, and does not depend on the estimator selected, $\widehat{m}$. The empirical quadratic risk, associated with a model $m$, is here:$$\widehat{\mathcal{R}}_n(m)=\frac{1}{n}\sum_{i=1}^n (y_i-m(\mathbf{x}_i))^2$$ (by convention). We recognize here the mean square error, “mse”, which will more generally give the “risk” of the model $m$ when using another loss function (as we will discuss later on). It should be noted that:$$\displaystyle{\mathbb{E}[\widehat{\mathcal{R}}_n(m)]=\frac{1}{n}\|m_0(\mathbf{x})-m(\mathbf{x})\|^2+\frac{1}{n}\mathbb{E}\big(\|{Y}-m_0(\mathbf{X})\|^2\big)}$$We can show that:$$n\mathbb{E}\big[\widehat{\mathcal{R}}_n(\widehat{m})\big]=\mathbb{E}\big(\|Y-\widehat{m}(\mathbf{x})\|^2\big)=\|(\mathbb{I}-\mathbf{S})m_0\|^2+\sigma^2\|\mathbb{I}-\mathbf{S}\|^2$$so that the (real) risk of $\widehat{m}$ is:$${\mathcal{R}}_n(\widehat{m})=\mathbb{E}\big[\widehat{\mathcal{R}}_n(\widehat{m})\big]+2\frac{\sigma^2}{n}\text{trace}(\boldsymbol{S})$$So, if $\text{trace}(\boldsymbol{S})\geq0$ (which is not a too strong assumption), the empirical risk underestimates the true risk of the estimator. Actually, we recognize here the number of degrees of freedom of the model, the right-hand term corresponding to Mallow’s $C_p$, introduced in Mallows (1973) using not deviance but $R^2$. ## Statistical Tests The most traditional test in econometrics is probably the significance test, corresponding to the nullity of a coefficient in a linear regression model. Formally, it is the test of $H_0:\beta_k=0$ against $H_1:\beta_k\neq 0$. The so-called Student test, based on the statistics $t_k=\widehat{\beta}_k/se_{\widehat{β}_k}$, allows to decide between the two alternatives, using the test $p$-value, defined by $\mathbb{P}[|T|>|t_k|]$ avec $T\overset{\mathcal{L}}{\sim} Std_\nu$, where $\nu$ is the number of degrees of freedom of the model ($\nu=p+1$ for the standard linear model). In large dimension, however, this statistic is of very limited interest, given a significant FDR (“False Discovery Ratio”). Classically, with a level of significance $\alpha=0.05$, 5% of the variables are falsely significant. Suppose that we have $p$=100 explanatory variables, but that 5 (only) are really significant. We can hope that these 5 variables will pass the Student test, but we can also expect that 5 additional variables (false positive test) will emerge. We will then have 10 variables perceived as significant, while only half are significant, i.e. an FDR ratio of 50%. In order to avoid this recurrent pitfall in multiple tests, it is natural to use the procedure of Benjamini & Hochberg (1995). ## From a correlation to some causal effect Econometric models are used to implement public policy evaluations. It is therefore essential to fully understand the underlying mechanisms in order to know which variables actually make it possible to act on a variable of interest. But then we move on to another important dimension of econometrics. Jerry Neyman was responsible for the first work on the identification of causal mechanisms, and then Rubin (1974) formalized the test, called the “Rubin causal model” in Holland (1986). The first approaches to the notion of causality in econometrics were based on the use of instrumental variables, models with discontinuity of regression, analysis of differences in differences, and natural or unnatural experiments. Causality is usually inferred by comparing the effect of a policy – or more generally of a treatment – with its counterfactual, ideally given by a random control group. The causal effect of the treatment is then defined as $\Delta=y_1-y_0$, i.e. the difference between what the situation would be with treatment (noted t=1) and without treatment (noted t=0). The concern is that only $y=t\cdot y_1+(1-t)\cdot y_0$ and $t$ are observed. In other words, the causal effect of variable $t$ on $t$ is not observed (since only one of the two potential variables – $y_0$ or $y_1$ is observed for each individual), but it is also individual, and therefore a function of x-covariates. Generally, by making assumptions about the distribution of the triplet $(Y_0,Y_1,T)$, some parameters of the causal effect distribution become identifiable, based on the density of the observable variables $(Y,T)$. Classically, we will be interested in the moments of this distribution, in particular the average effect of treatment in the population, $\mathbb{E}[\Delta]$, or even just the average effect of treatment in the case of treatment $\mathbb{E}[\Delta|T=1]$. If the result $(Y_0,Y_1)$ is independent of the processing access variable $T$, it can be shown that $\mathbb{E}[\Delta]=\mathbb{E}[Y|T=1]- \mathbb{E} [Y|T=0]$. But if this independence hypothesis is not verified, there is a selection bias, often associated with $\mathbb{E}[Y_0|T=1]- \mathbb{E} [Y_0|T=0]$. Rosenbaum & Rubin (1983) propose to use a propensity to be treated score, $p(x)=\mathbb{P}[T=1|X=x]$, noting that if variable $Y_0$\ is independent of access to treatment $T$ conditionally to the explanatory variables $X$, then it is independent of $T$ conditionally to the score $p(X)$ : it is sufficient to match them using their propensity score. Heckman et al (2003) thus proposes a kernel estimator on the propensity score, which simply provides an estimator of the effect of the treatment, provided that it is treated. To be continued next time, we’ll introduce “machine learning techniques” (references mentioned above are online here) # Probabilistic Foundations of Econometrics, part 3 This post is the third one of our series on the history and foundations of econometric and machine learning models. Part 2 is online here. ## Exponential family and linear models The Gaussian linear model is a special case of a large family of linear models, obtained when the conditional distribution of $Y$ (given the covariates) belongs to the exponential family$$f(y_i|\theta_i,\phi)=\exp\left(\frac{y_i\theta_i-b(\theta_i)}{a(\phi)}+c(y_i,\phi)\right)$$ with $\theta_i=\psi(\mathbf{x}_i^T \beta)$. Functions $a$, $b$ and $c$ are specified according to the type of exponential law (studied extensively in statistics since Darmoix (1935), as Brown (1986) reminds us), and $\psi$ is a one-to-one mapping that the user must specify. Log-likelihood then has a simple expression $$\log\mathcal{L}(\mathbf{\theta},\phi|\mathbf{y}) =\frac{\sum_{i=1}^ny_i\theta_i-\sum_{i=1}^nb(\theta_i)}{a(\phi)}+\sum_{i=1}^n c(y_i,\phi)$$ and the first order condition is then written$$\frac{\partial \log \mathcal{L}(\mathbf{\theta},\phi|\mathbf{y})}{\partial \mathbf{\beta}} = \mathbf{X}^T\mathbf{W}^{-1}[\mathbf{y}-\widehat{\mathbf{y}}]=\mathbf{0}$$based on Müller’s (2011) notations, where $\mathbf{W}$ is a weight matrix (which depends on $\beta$). Given the link between $\theta$ and the expectation of $Y$, instead of specifying the function $\psi(\cdot)$, we will tend to specify the link function $g(\cdot)$ defined by $$\widehat{y}=m(\mathbf{x})=\mathbb{E}[Y|\mathbf{X}=\mathbf{x}]=g^{-1} (\mathbf{x}^T \beta)$$For the Gaussian linear regression we consider an identity link, while for the Poisson regression, the natural link (called canonical) is the logarithmic link. Here, as $\mathbf{W}$ depends on $\beta$ (with $\mathbf{W}=diag(\nabla g(\widehat{\mathbf{y}})Var[\mathbf{y}])$ there is generally no explicit formula for the maximum likelihood estimator. But an iterative algorithm makes it possible to obtain a numerical approximation. By setting $$\mathbf{z}=g(\widehat{\mathbf{y}})+(\mathbf{y}-\widehat{\mathbf{y}})\cdot\nabla g(\widehat{\mathbf{y}})$$ corresponding to the error term of a Taylor development in order 1 of $g$, we obtain an algorithm of the form$$\widehat{\beta}_{k+1}=[\mathbf{X}^T \mathbf{W}_k^{-1} \mathbf{X}]^{-1} \mathbf{X}^T \mathbf{W}_k^{-1} \mathbf{z}_k$$By iterating, we will define $\widehat{\beta}=\widehat{\beta}_{\infty}$, and we can show that – with some additional technical assumptions (detailed in Müller (2011)) – this estimator is asymptotically Gaussian, with $$\sqrt{n}(\widehat{\beta} -\beta)\overset{\mathcal{L}}{\rightarrow} \mathcal{N}(\mathbf{0},I(β)^{-1})$$where numerically $I(\beta)=\varphi\cdot[\mathbf{X}^T \mathbf{W}_\infty^{-1} \mathbf{X}]$. From a numerical point of view, the computer will solve the first-order condition, and actually, the law of $Y$ does not really intervene. For example, one can estimate a “Poisson regression” even when observations are not integers (but they need to be positive). In other words, the law of $Y$ is only an interpretation here, and the algorithm could be introduced in a different way (as we will see later on), without necessarily having an underlying probabilistic model. ## Logistic Regression Logistic regression is the generalized linear model obtained with a Bernoulli’s law, and a link function which is the quantile function of a logistic law (which corresponds to the canonical link in the sense of the exponential family). Taking into account the form of Bernoulli’s law, econometrics proposes a model for $y_i\in\{0,1\}$, in which the logarithm of the odds follows a linear model: $$\log\left(\frac{\mathbb{P}[Y=1\vert \mathbf{X}=\mathbf{x}]}{\mathbb{P}[Y\neq 1\vert \mathbf{X}=\mathbf{x}]}\right)=\beta_0+\mathbf{x}^T\beta$$or $$\mathbb{E}[Y|\mathbf{X}=\mathbf{x}]=\mathbb{P}[Y=1|\mathbf{X}=\mathbf{x}]=\frac{e^{\beta_0+\mathbf{x}^T\beta}}{1+ e^{\beta_0+\mathbf{x}^T\beta}}=H(\beta_0+\mathbf{x}^T\beta)$$where $H(\cdot)=\exp(\cdot)/(1+exp(\cdot))$ is the cumulative distribution function of the logistic law. The estimation of $(\beta_0,\beta)$ is performed by maximizing the likelihood: $$\mathcal{L}=\prod_{i=1}^n \left(\frac{e^{\mathbf{x}_i^T\mathbf{\beta}}}{1+e^{\boldsymbol{x}_i^T\mathbf{\beta}}}\right)^{y_i}\left(\frac{1}{1+e^{\mathbf{x}_i^T\mathbf{\beta}}}\right)^{1-y_i}$$ It is said to be a linear models because isoprobability curves here are the parallel hyperplanes $b+\mathbf{x}^T\beta$. Rather than this model, popularized by Berkson (1944), some will prefer the probit model (see Berkson, 1951), introduced by Bliss (1934). In this model: $$\mathbb{E}[Y|\mathbf{X}=\mathbf{x}]=\mathbb{P}[Y=1|\mathbf{X}=\mathbf{x}]=\Phi (\beta_0+\mathbf{x}^T\beta)$$ where $\Phi$ denotes the distribution function of the reduced centred normal distribution. This model has the advantage of having a direct link with the Gaussian linear model, since $y_i=\mathbf{1}(y_i^\star>0)$ with $y_i^\star=\beta_0+\mathbf{x}^T \beta+\varepsilon_i$ where the residuals are Gaussian, $\mathcal{N}(0,\sigma^2)$. An alternative is to have centered residuals of unit variance, and to consider a latent modeling of the form $y_i=\mathbf{1}(y_i^\star>\xi)$ (where $\xi$ will be fixed). As we can see, these techniques are fundamentally linked to an underlying stochastic model. In the body of the article, we present several alternative techniques – from the learning literature – for this classification problem (with two classes, here $0$ and $1$). ## Regression in high dimension As we mentioned earlier, the first order condition $\mathbf{X}^T (\mathbf{X}\widehat{\beta}-\mathbf{y})=\mathbf{0}$ is solved numerically by performing a QR decomposition, at a cost which consists in $O(np^2)$ operations (where $p$ is the rank of $\mathbf{X}^T \mathbf{X}$). Numerically, this calculation can be long (either because $p$ is large or because $n$ is large), and a simpler strategy may be to sub-sample. Let $n_s\ll n$, and consider a sub-sample size $n_s$ of $\{1,\cdots,n\}$. Then $\widehat{\beta}_s=(\mathbf{X}_s^T \mathbf{X}_s )^{-1} \mathbf{X}_s^T\mathbf{y}_s$ is a good approximation of $\beta$ as shown by Dhillon et al. (2014). However, this algorithm is dangerous if some points have a high leverage (i.e. $L_i=\mathbf{x}_i(\mathbf{X}^T\mathbf{X})^{-1}\mathbf{x}_i^T$). Tropp (2011) proposes to transform the data (in a linear way), but a more popular approach is to do non-uniform sub-sampling, with a probability related to the influence of observations (defined by $I_i=\widehat{\varepsilon}_iL_i/(1-L_i)^2$, and which unfortunately can only be calculated once the model is estimated). In general, we will talk about massive data when the data table of size does not fit in the RAM memory of the computer. This situation is often encountered in statistical learning nowadays with very often $p\ll n$. This is why, in practice, many libraries of algorithms assimilated to machine learning use iterative methods to solve the first-order condition. When the parametric model to be calibrated is indeed convex and semi-differentiable, it is possible to use, for example, the stochastic gradient descent method as suggested by Bottou (2010). This last one allows to free oneself at each iteration from the calculation of the gradient on each observation of our learning base. Rather than making an average descent at each iteration, we start by drawing (without replacement) an observation $\mathbf{x}_i$ among the $n$ available. The model parameters are then corrected so that the prediction made from $\mathbf{x}_i$ is as close as possible to the true value $y_i$. The method is then repeated until all the data have been reviewed. In this algorithm there is therefore as much iteration as there are observations. Unlike the gradient descent algorithm (or Newton’s method) at each iteration, only one gradient vector is calculated (and no longer $n$). However, it is sometimes necessary to run this algorithm several times to increase the convergence of the model parameters. If the objective is, for example, to minimize a loss function $\ell$ between the estimator $m_\beta (\mathbf{x})$ and $y$ (like the quadratic loss function, as in the Gaussian linear regression) the algorithm can be summarized as follows: • Step 0: Mix the data • Iteration step: For $t=1,\cdots, n$, we pull $i\in\{1,\cdots,n\}$ without replacement, and we set $$\beta^{t+1} = \beta^{t} - \gamma_t\frac{ \partial{\ell(y_i,m_{\beta^t}(X_i)) } }{ \partial{ \beta}}$$ This algorithm can be repeated several times as a whole depending on the user’s needs. The advantage of this method is that at each iteration, it is not necessary to calculate the gradient on all observations (more sum). It is therefore suitable for large databases. This algorithm is based on a convergence in probability towards a neighborhood of the optimum (and not the optimum itself). (references will be given in the very last post of that series) To be continued # Probabilistic Foundations of Econometrics, part 2 This post is the second one of our series on the history and foundations of econometric and machine learning models. Part 1 is online here. ## Geometric Properties of this Linear Model Let’s define the scalar product in $\mathbb{R}^n$, $⟨\mathbf{a},\mathbf{b}⟩=\mathbf{a}^T\mathbf{b}$, and let’s note $\|\cdot\|$ the associated Euclidean standard, $\|\mathbf{a}\|=\sqrt{\mathbf{a}^T\mathbf{a}}$ (denoted $\|\cdot\|_{\ell_2}$ in the next post). Note $\mathcal{E}_X$ the space generated by all linear combinations of the $\mathbf{X}$ components (adding the constant). If the explanatory variables are linearly independent, $\mathbf{X}$ is a full (column) rank matrix and $\mathcal{E}_X$ is a space of dimension $p+1$. Let’s assume from now on that the variables $\mathbf{x}$ and $y$ are centered here. Note that no law hypothesis is made in this section, the geometric properties are derived from the properties of expectation and variance in the set of finite variance variables. With this notation, it should be noted that the linear model is written $m(\mathbf{x})=⟨\mathbf{x},\beta⟩$. The space $H_z=\{\mathbf{x}\in\mathbb{R}^{p+1}:m(\mathbf{x})=z\}$ is a hyperplane (affine) that separates the space in two. Let’s define the orthogonal projection operator on $\mathcal{E}_X$, $\Pi_X =\mathbf{X}(\mathbf{X}^T\mathbf{X})^{-1} \mathbf{X}^T$. Thus, the forecast that can be made for it is: $$\widehat{\mathbf{y}}=\mathbf{X}(\mathbf{X}^T\mathbf{X})^{-1} \mathbf{X}^T\mathbf{y}=\Pi_X\mathbf{y}$$. As, $\widehat{\varepsilon}=\mathbf{y}-\widehat{\mathbf{y}}=(\mathbb{I}-\Pi_X)\mathbf{y}=\Pi_{X^\perp}\mathbf{y}$, we note that $\widehat{\varepsilon}\perp\mathbf{x}$, which will be interpreted as meaning that residuals are a term of innovation, unpredictable in the sense that $\Pi_{X }\widehat{\varepsilon}=\mathbf{0}$. The Pythagorean theorem is written here: $$\Vert \mathbf{y} \Vert^2=\Vert \Pi_{ {X}}\mathbf{y} \Vert^2+\Vert \Pi_{ {X}^\perp}\mathbf{y} \Vert^2=\Vert \Pi_{ {X}}\mathbf{y}\Vert^2+\Vert \mathbf{y}-\Pi_{ {X}}\mathbf{y}\Vert^2=\Vert\widehat{\mathbf{y}}\Vert^2+\Vert\widehat{\mathbf{\varepsilon}}\Vert^2$$which is classically translated in terms of the sum of squares: $$\underbrace{\sum_{i=1}^n y_i^2}_{n\times\text{total variance}}=\underbrace{\sum_{i=1}^n \widehat{y}_i^2}_{n\times\text{explained variance}}+\underbrace{\sum_{i=1}^n (y_i-\widehat{y}_i)^2}_{n\times\text{residual variance}}$$The coefficient of determination, $R^2$, is then interpreted as the square of the cosine of the angle $\theta$ between $\mathbf{y}$ and $\Pi_X \mathbf{y}$ : $$R^2=\frac{\Vert \Pi_{{X}} \mathbf{y}\Vert^2}{\Vert \mathbf{y}\Vert^2}=1-\frac{\Vert \Pi_{ {X}^\perp} \mathbf{y}\Vert^2}{\Vert \mathbf {y}\Vert^2}=\cos^2(\theta)$$An important application was obtained by Frish & Waugh (1933), when the explanatory variables are divided into two groups, $\mathbf{X}=[\mathbf{X}_1 |\mathbf{X}_2]$, so that the regression becomes $y=\beta_0+\mathbf{X}_1 β_1+\mathbf{X}_2 β_2+\varepsilon$. Frish & Waugh (1933) showed that two successive projections could be considered. Indeed, if $\mathbf{y}_2^\star=\Pi_{X_1^\perp} \mathbf{y}$ and $X_2^\star=\Pi_{X_1^\perp}\mathbf{X}_2$, we can show that $$\widehat{\beta} _2=[{\mathbf{X}_2^\star}^T \mathbf{X}_2^\star]^{-1}{\mathbf{X}_2^\star}^T \mathbf{y}_2^\star$$ In other words, the overall estimate is equivalent to the combination of independent estimates of the two models if $\mathbf{X}_2^\star=\mathbf{X}_2$, i.e. $\mathbf{X}_2\in \mathcal{E}_{X_1}^\perp$, which can be noted $\mathbf{x}_1\perp\mathbf{x}_2$ We obtain here the Frisch-Waugh theorem which guarantees that if the explanatory variables between the two groups are orthogonal, then the overall estimate is equivalent to two independent regressions, on each of the sets of explanatory variables. This is a theorem of double projection, on orthogonal spaces. Many results and interpretations are obtained through geometric interpretations (fundamentally related to the links between conditional expectation and the orthogonal projection in space of variables of finite variance). This geometric interpretation might help to get a better understanding of the problem of under-identification, i.e. the case where the real model would be $y_i=\beta_0+ \mathbf{x}_1^T \beta_1+\mathbf{x}_2^T \beta_2+\varepsilon_i$, but the estimated model is $y_i=b_0+\mathbf{x}_1^T \mathbf{b}_1+\eta_i$. The maximum likelihood estimator of $\mathbf{b}_1$ is $$\widehat{\mathbf{b}}_1=\mathbf {\beta}_1 + \underbrace{ (\mathbf {X}_1^T\mathbf {X}_1)^{-1} \mathbf {X}_1^T \mathbf {X}_{2} \mathbf{\beta}_2}_{\mathbf{\beta}_{12}}+\underbrace{(\mathbf{X}_1^{T}\mathbf{X}_1)^{-1} \mathbf{X}_1^T\varepsilon}_{\nu}$$so that $\mathbb{E}[\widehat{\mathbf{b}}_1]=\beta_1+\beta_{12}$, the bias ($\beta_{12}$) being null only in the case where $\mathbf{X}_1^T \mathbf{X}_2=\mathbf{0}$ (i. e. $\mathbf{X}_1\perp \mathbf{X}_2$): we find here a consequence of the Frisch-Waugh theorem. On the other hand, over-identification corresponds to the case where the real model would be $y_i=\beta_0+\mathbf{x}_1^T \beta_1+\varepsilon_i$, but the estimated model is $y_i=b_0+ \mathbf{x}_1^T \mathbf{b} _1+\mathbf{x}_2^T \mathbf{b}_2+\eta_i$. In this case, the estimate is unbiased, in the sense that $\mathbb{E}[\widehat{\mathbf{b}}_1]=\beta_1$ but the estimator is not efficient. Later on, we will discuss an effective method for selecting variables (and avoid over-identification). ## From parametric to non-parametric We can rewrite equation (4) in the form $\widehat{\mathbf{y}}=\Pi_X\mathbf{y}$ which helps us to see the forecast directly as a linear transformation of the observations. More generally, a linear predictor can be obtained by considering $m(\mathbf{x})=\mathbf{s}_{\mathbf{x}}^T \mathbf{y}$, where $\mathbf{s}_{\mathbf{x}}$ is a weight vector, which depends on $\mathbf{x}$, interpreted as a smoothing vector. Using the vectors $\mathbf{s}_{\mathbf{x}_i}$, calculated from the observations $\mathbf{x}_i$, we obtain a matrix $\mathbf{S}$ of size $n\times n$, and $\widehat{\mathbf{y}}=\mathbf{S}\mathbf{y}$. In the case of the linear regression described above, $\mathbf{s}_{\mathbf{x}}=\mathbf{X}[\mathbf{X}^T\mathbf{X}]^{-1}\mathbf{x}$, and in that case $\text{trace}(\mathbf{S})$ is the number of columns in the $\mathbf{X}$ matrix (the number of explanatory variables). In this context of more general linear predictors, $\text{trace}(\mathbf{S})$ is often seen as equivalent to the number of parameters (or complexity, or dimension, of the model), and $\nu=n-\text{trace}(\mathbf{S})$ is then the number of degrees of freedom (see Ruppert et al., 2003; Simonoff, 1996). The principle of parsimony says that we should minimize this dimension (the trace of the matrix $\mathbf{S}$) as much as possible. But in the general case, this dimension is more to obtain, explicitely. The estimator introduced by Nadaraya (1964) and Watson (1964), in the case of a simple non-parametric regression, is also written in this form since$$\widehat{m}_h(x)=\mathbf{s}_{x}^T\mathbf{y}=\sum_{i=1}^n \mathbf{s}_{x,i}y_i$$where$$\mathbf{s}_{x,i}=\frac{K_h(x-x_i)}{K_h(x-x_1)+\cdots+K_h(x-x_n)}$$ where $K(\cdot)$ is a kernel function, which assigns a value that is lower the closer $x_i$ is to $x$, and $h>0$ is the bandwidth. The introduction of this metaparameter $h$ is an important issue, as it should be chosen wisely. Using asymptotic developments, we can show that if $X$ has density $f$, $$\text{biais}[\widehat{m}_h(x)]=\mathbb{E}[\widehat{m}_h(x)]-m(x)\sim {h^2}\left(\frac{C_1 }{2}m''(x)+C_2 m'(x)\frac{f'(x)}{f(x)}\right)$$and $$\displaystyle{{\text{Var}[\widehat{m}_h(x)]\sim\frac{C_3}{{nh}}\frac{\sigma(x)}{f(x)}}}$$for some constants that can be estimated (see Simonoff (1996) for a discussion). These two functions evolve inversely with $h$, as shown in Figure 1 (where the metaparameter on the $x$-axis is here, actually, $h^{-1}$). Keep in ming that we will see a similar graph in the context of machine learning models. Figure 1. Choice of meta-parameter and the Goldilocks problem: it must not be too large (otherwise there is too much variance), nor too small (otherwise there is too much bias). The natural idea is then to try to minimize the mean square error, the MSE, defined as $bias[\widehat{m}_h (x)]^2+Var[\widehat{m}_h (x)]$, and them integrate over $x$, which gives an optimal value for $h$ of the form $h^\star=O(n^{-1/5})$, and reminds us of Silverman’s rule – see Silverman (1986). In larger dimensions, for continuous $\mathbf{x}$ variables, a multivariate kernel with matrix bandwidth $\mathbf{H}$ can be used, and$$\mathbb{E}[\widehat{m}_{\mathbf{H}}(\mathbf{x})]\sim m(\mathbf{x})+\frac{C_1}{2}\text{trace}\big(\mathbf{H}^Tm''(\mathbf{x})\mathbf{H}\big)+C_2\frac{m'(\boldsymbol{x})^T\mathbf{H}\mathbf{H}^T \nabla f(\mathbf{x})}{f(\mathbf{x})}$$while$$\text{Var}[\widehat{m}_{\mathbf{H}}(\mathbf{x})]\sim\frac{C_3}{n~\text{det}(\mathbf{H})}\frac{\sigma(\mathbf{x})}{f(\mathbf{x})}$$ If $\mathbf{H}$ is a diagonal matrix, with the same term $h$ on the diagonal, then $h^\star=O(n^{-1/(4+dim(\mathbf{x}))}$. However, in practice, there will be more interest in the integrated version of the quadratic error, $$MISE(\widehat{m}_{h})=\mathbb{E}[MSE(\widehat{m}_{h}(X))]=\int MSE(\widehat{m}_{h}(x))dF(x)$$and we can prove that $$MISE[\widehat{m}_h]\sim \overbrace{\frac{h^4}{4}\left(\int x^2k(x)dx\right)^2\int\big[m''(x)+2m'(x)\frac{f'(x)}{f(x)}\big]^2dx}^{\text{bias}^2} +\overbrace{\frac{\sigma^2}{nh}\int k^2(x)dx \cdot\int\frac{dx}{f(x)}}^{\text{variance}}$$as n→∞ and nh→∞. Here we find an asymptotic relationship that again recalls Silverman’s (1986) order of magnitude, $$h^\star =n^{-\frac{1}{5}}\left(\frac{C_1\int \frac{dx}{f(x)}}{C_2\int \big[m''(x)+2m'(x)\frac{f'(x)}{f(x)}\big]dx}\right)^{\frac{1}{5}}$$The main problem here, in practice, is that many of the terms in the expression above are unknown. Automatic learning offers computational techniques, when the econometrician used to searching for asymptotic (mathematical) properties. To be continued (references mentioned above are online here)… # Probabilistic Foundations of Econometrics, part 1 In a series of posts, I wanted to get into details of the history and foundations of econometric and machine learning models. It will be some sort of online version of our joint paper with Emmanuel Flachaire and Antoine Ly, Econometrics and Machine Learning (initially writen in French), that will actually appear soon in the journal Economics and Statistics. This is the first one… The importance of probabilistic models in economics is rooted in Working’s (1927) questions and the attempts to answer them in Tinbergen’s two volumes (1939). The latter have subsequently generated a great deal of work, as recalled by Duo (1993) in his book on the foundations of econometrics, and more particularly in the first chapter “The Probability Foundations of Econometrics”. It should be recalled that Trygve Haavelmo was awarded the Nobel Prize in Economics in 1989 for his “clarification of the foundations of the probabilistic theory of econometrics”. Because as Haavelmo (1944) (initiating a profound change in econometric theory in the 1930s, as recalled in Morgan’s Chapter 8 (1990)) showed, econometrics is fundamentally based on a probabilistic model, for two main reasons. First, the use of statistical quantities (or “measures”) such as means, standard errors and correlation coefficients for inferential purposes can only be justified if the process generating the data can be expressed in terms of a probabilistic model. Second, the probability approach is relatively general, and is particularly well suited to the analysis of “dependent” and “non-homogeneous” observations, as they are often found on economic data.We will then assume that there is a probabilistic space $(\Omega,\mathcal{F},\mathbb{P})$ such that observations $(y_i,\mathbf{x}_i)$ are seen as realizations of random variables $(Y_i, \mathbf{X}_i)$. In practice, however, we are not very interested in the joint law of the couple $(Y, \mathbf{X})$ : the law of $\mathbf{X}$ is unknown, and it is the law of Y conditional on $\mathbf{X}$ that will be interested in. In the following, we will note $x$ a single observation, $\mathbf{x}$ a vector of observations, $X$ a random variable, and $\mathbf{X}$ a random vector. Abusively, $\mathbf{X}$ may also designate the matrix of individual observations (denoted $\mathbf{x}_i$), depending on the context. ## Foundations of mathematical statistics As recalled in Vapnik’s (1998) introduction, inference in parametric statistics is based on the following belief: the statistician knows the problem to be analyzed well, in particular, he knows the physical law that generates the stochastic properties of the data, and the function to be found is written via a finite number of parameters[1]. To find these parameters, the maximum likelihood method is used. The purpose of the theory is to justify this approach (by discovering and describing its favorable properties). We will see that in learning, philosophy is very different, since we do not have a priori reliable information on the statistical law underlying the problem, nor even on the function we would like to approach (we will then propose methods to construct an approximation from the data at our disposal, as in (1998)). A “golden age” of parametric inference, from 1930 to 1960, laid the foundations for mathematical statistics, which can be found in all statistical textbooks, including today. As Vapnik (1998) states, the classical parametric paradigm is based on the following three beliefs: 1. To find a functional relationship from the data, the statistician is able to define a set of functions, linear in their parameters, that contain a good approximation of the desired function. The number of parameters describing this set is small. 2. The statistical law underlying the stochastic component of most real-life problems is the normal law. This belief has been supported by reference to the central limit theorem, which stipulates that under large conditions the sum of a large number of random variables is approximated by the normal law. 3. The maximum likelihood method is a good tool for estimating parameters. In this section we will come back to the construction of the econometric paradigm, directly inspired by that of classical inferential statistics. ## Conditional laws and likelihood Linear econometrics has been constructed under the assumption of individual data, which amounts to assuming independent variables $(Y_i, \mathbf{X}_i)$ (if it is possible to imagine temporal observations – then we would have a process $(Y_t, \mathbf{X}_t)$ – but we will not discuss time series here). More precisely, we will assume that, conditionally to the explanatory variables $\mathbf{X}_i$, the variables $Y_i$ are independent. We will also assume that these conditional laws remain in the same parametric family, but that the parameter is a function of $\mathbf{x}$. In the Gaussian linear model it is assumed that: $$(Y\vert \mathbf{X}=\mathbf{x})\overset{\mathcal{L}}{\sim}\mathcal{N}(\mu(\mathbf{x}),\sigma^2)~~~~ (1)$$where $\mu(\mathbf{x})=\beta_0+\mathbf{x}^T\mathbf{\beta}$ and $\mathbf{\beta}\in\mathbb{R}^{p}$. It is usually called a ‘linear’ model since $\mathbb{E}[Y\vert \mathbf{X}=\mathbf{x}]=\beta_0+\mathbf{x}^T\mathbf{\beta}$ is a linear combination of covariates[2]. It is said to be a homoscedastic model if $Var[Y|\mathbf{X}=\mathbf{x}]=\sigma^2$, where $\sigma^2$ is a positive constant. To estimate the parameters, the traditional approach is to use the Maximum Likelihood estimator, as initially suggested by Ronald Fisher. In the case of the Gaussian linear model, log-likelihood is written: $$\log\mathcal{L}(\beta_0, \mathbf{\beta},\sigma^2\vert \mathbf{y},\mathbf{x}) = -\frac{n}{2}\log[2\pi\sigma^2] - \frac{1}{2\sigma^2}\sum_{i=1}^n (y_i-\beta_0-\mathbf{x}_i^T\mathbf{\beta})^2$$Note that the term on the right, measuring a distance between the data and the model, will be interpreted as deviance in generalized linear models. Then we will set: $$(\widehat{\beta}_0,\widehat{\mathbf{\beta}},\widehat{\sigma}^2)=\text{argmax}\left\lbrace\log\mathcal{L}(\beta_0, \mathbf{\beta},\sigma^2\vert \mathbf{y},\mathbf{x})\right\rbrace$$The maximum likelihood estimator is obtained by minimizing the sum of the error squares (the so-called “least squares” estimator) that we will find in the “machine learning” approach. The first order conditions allow to find the normal equations, whose matrix writing is $\mathbf{X}^T[\mathbf{y}-\mathbf{X}\mathbf{\beta}]=\mathbf{0}$, which can also be written $(\mathbf{X}^T \mathbf{X})\mathbf{\beta}=\mathbf{X}^T \mathbf{y}$. If $\mathbf{X}$ is a full (column) rank matrix, then we find the classical estimator:$$\widehat{\mathbf{\beta}}=(\mathbf{X}^T\mathbf{X})^{-1}\mathbf{X}^T\mathbf{y}=\mathbf{\beta}+(\mathbf{X}^T\mathbf{X})^{-1}\mathbf{X}^{-1}\mathbf{\varepsilon}~~~(2)$$using residual-based writing (as often in econometrics), $y=\mathbf{x}^T\mathbf{\beta}+\varepsilon$. Gauss Markov’s theorem ensures that this estimator is the unbiased linear estimator with minimum variance. It can then be shown that $\widehat{\mathbf{\beta}}\sim\mathcal{N}(\mathbf{\beta},\sigma^2(\mathbf{X}^T\mathbf{X})^{-1})$, and in particular, if we simply need the first two moments : $$\mathbb{E}[\widehat{\mathbf{\beta}}]=\mathbf{\beta}~~~Var[\widehat{\mathbf{\beta}}]=\sigma^2 [\mathbf{X}^T\mathbf{X}]^{-1}$$In fact, the normality hypothesis makes it possible to make a link with mathematical statistics, but it is possible to construct this estimator given by equation (2) without that Gaussian assumption. Hence, if we assume that $Y|\mathbf{X}$ has the same distribution as $\mathbf{x}^T\mathbf{\beta}+\varepsilon$, where $\mathbb{E}[\varepsilon]=0$, $Var[\varepsilon]=\sigma^2$ and $Cov[X_j,\varepsilon]=0$ for all $j$, then $\widehat{\mathbf{\beta}}$ is an unbiased estimator of $\mathbf{\beta}$ with smallest variance[3] among unbiased linear estimators. Furthermore, if we cannot get normality at finite distance, asymptotically this estimator is Gaussian, with $$\sqrt{n}(\widehat{\mathbf{\beta}}-\mathbf{\beta})\overset{\mathcal{L}}{\rightarrow}\mathcal{N}(\mathbf{0},\mathbf{\Sigma})$$as $n\rightarrow\infty$, for some matrix $\mathbf{\Sigma}$. The condition of having a full rank $\mathbf{X}$ matrix can be (numerically) strong in large dimensions. If it is not satisfied, $(\mathbf{X}^T \mathbf{X})^{-1}\mathbf{X}^T$ does not exist. If $\mathbb{I}$ denotes the identity matrix, however, it should be noted that $(\mathbf{X}^T \mathbf{X}+\lambda\mathbb{I})^{-1}\mathbf{X}^T$ still exists, whatever $\lambda>0$. This estimator is called the ridge estimator of level \lambda (introduced in the 1960s by Hoerl (1962), and associated with a regularization studied by Tikhonov (1963)). This estimator naturally appears in a Bayesian econometric context. ## Residuals It is not uncommon to introduce the linear model from the distribution of the residuals, as we mentioned earlier. Also, equation (1) is written as often: $$y_i=\beta_0+\mathbf{x}_i^T\mathbf{\beta}+\varepsilon_i~~~~(3)$$where $\varepsilon_i$’s are realizations of independent and identically distributed random variables (i.i.d.) from some $\mathcal{N}(0,\sigma^2)$ distribution. With a vector notation, we will write $\mathbf{\varepsilon}\overset{\mathcal{L}}{\sim}\mathcal{N}(\mathbf{0},\sigma^2\mathbb{I})$. The estimated residuals are defined as: $$\widehat{\varepsilon}_i =y_i-[\widehat{\beta}_0+\mathbf{x}_i^T\widehat{\mathbf{\beta}}]$$ Those (estimated) residuals are basic tools for diagnosing the relevance of the model. An extension of the model described by equation (1) has been proposed to take into account a possible heteroscedastic character: $$(Y\vert \mathbf{X}=\mathbf{x})\overset{\mathcal{L}}{\sim}\mathcal{N}(\mu(\mathbf{x}),\sigma^2(\mathbf{x}))$$where $\sigma^2(\mathbf{x})$ is a positive function of the explanatory variables. This model can be rewritten as: $$y_i=\beta_0+\mathbf{x}_i^T\mathbf{\beta}+\sigma^2(\mathbf{x}_i)\cdot\varepsilon_i$$where residuals are always i.i.d., with unit variance, $$\varepsilon_i=\frac{y_i-[\beta_0+\mathbf{x}_i^T\mathbf{\beta}]}{\sigma(\mathbf{x}_i)}$$ While residuals based equations are popular in linear econometrics (when the dependent variable is continuous), it is no longer popular in counting models, or logistic regression. However, writing using an error term (as in equation (3)) raises many questions about the representation of an economic relationship between two quantities. For example, it can be assumed that there is a relationship (linear to begin with) between the quantities of a traded good, $q$ and its price $p$. This allows us to imagine a supply equation$$q_i=\beta_0+\beta_1 p_i+u_i$$($u_i$ being an error term) where the quantity sold depends on the price, but in an equally legitimate way, one can imagine that the price depends on the quantity produced (what one could call a demand equation), $$p_i=\alpha_0+\alpha_1 q_i+v_i$$($v_i$ denoting another error term). Historically, the error term in equation (3) could be interpreted as an idiosyncratic error on the variable $y$, the so-called explanatory variables being assumed to be fixed, but this interpretation often makes the link between an economic relationship and a complicated economic model difficult, the economic theory speaking abstractly about a relationship between a magnitude, the econometric model imposing a specific shape (what magnitude is $y$ and what magnitude is $x$) as shown in more detail in Morgan (1990) Chapter 7. (references mentioned above are online here). To be continued… [1] This approach can be compared to structural econometrics, as presented for example in Kean (2010). [2] Here, we will try to distinguish $\beta_0$, the intercept, and the other parameters $\mathbf{\beta}$, since they are considered differently in many extensions (e.g. regularization). Nevertheless, in many expressions $\mathbf{\beta}$ will denote the joint vector $(\beta_0, \mathbf{\beta})$, for general formulas, to avoid too heavy notations. [3] In the sense that the difference between variance matrices is a positive matrix. # On the robustness of LASSO Probably the last post on lasso, before the summer break… More specifically, I was wondering about the interpretation of graphs $\lambda\mapsto\widehat{\beta}_\lambda$. We use them for variable selection, but my major concern was about confidence intervals : how can we trust those lines ? As usual, a natural way is to use simulations on generated datasets. Consider for instance Sigma = matrix(c(1,.8,.2,.8,1,.4,.2,.4,1),3,3) n = 1000 library(mnormt) X = rmnorm(n,rep(0,3),Sigma) set.seed(123) df = data.frame(X1=X[,1],X2=X[,2],X3=X[,3],X4=rnorm(n), X5=runif(n), X6=exp(X[,3]), X7=sample(c("A","B"),size=n,replace=TRUE,prob=c(.5,.5)), X8=sample(c("C","D"),size=n,replace=TRUE,prob=c(.5,.5))) df$Y = 1+df$X1-df$X4+5*(df$X7=="A")+rnorm(n) One can use other simulations of datasets, and store the output vlambda = exp(seq(-8,1,length=201)) lasso = glmnet(x=X,y=df[,"Y"],family="gaussian",alpha=1, lambda=vlambda,standardize=TRUE) VLASSO[[s]] = as.matrix(lasso$beta)

To visualize confidence bands, one can compute quantiles

Q05=Q95=Qm=matrix(NA,9,201) for(i in 1:nrow(Q05)){ for(j in 1:ncol(Q05)){ v = unlist(lapply(VLASSO,function(x) x[i,j])) Q05[i,j] = quantile(v,.05) Q95[i,j] = quantile(v,.95) Qm[i,j] = mean(v) }}

and get get the graph

plot(lasso,col=colrs,"lambda"ylim=c(min(Q05),max(Q95))) colrs=c(brewer.pal(8,"Set1")) polygon(c(log(lasso$lambda),rev(log(lasso$lambda))), c(Q05[2,],rev(Q95[2,])),col=colrs[1],border=NA) polygon(c(log(lasso$lambda),rev(log(lasso$lambda))), c(Q05[5,],rev(Q95[5,])),col=colrs[2],border=NA) polygon(c(log(lasso$lambda),rev(log(lasso$lambda))), c(Q05[8,],rev(Q95[8,])),col=colrs[3],border=NA)

An alternative (more realistic on real data) is to use bootstrapped version of the dataset

id = sample(1:nrow(X),size=nrow(X),replace=TRUE) lasso = glmnet(x=X[id,],y=df[id,"Y"],family="gaussian",alpha=1, lambda=vlambda,standardize=TRUE)

So far, it looks it’s working very well. Now, what if we have a smaller dataset

n = 100

On simulated new samples, we get

while the bootstrap version is

There is more uncertainty, clearly, but the conclusion is not ambiguous here.

Now, what about real data. Consider the following

chicago = read.table("http://freakonometrics.free.fr/chicago.txt",header=TRUE,sep=";") tail(chicago) Fire X_1 X_2 X_3 42 4.8 0.152 19 13.323 43 10.4 0.408 25 12.960 44 15.6 0.578 28 11.260 45 7.0 0.114 3 10.080 46 7.1 0.492 23 11.428 47 4.9 0.466 27 13.731

with one variable of interest (the number of fires, per unhabitants) and 3 features. We can here use bootstrap to generate samples, and then fit a lasso regression. On the original dataset, the regression is

X = model.matrix(lm(Fire~.,data=chicago)) id = sample(1:nrow(X),size=nrow(X),replace=TRUE) vlambda = exp(seq(-4,2,length=201)) lasso = glmnet(x=X[id,],y=chicago[id,"Fire"],family="gaussian",alpha=1, lambda=vlambda,standardize=TRUE)

And if we just plot lines $\lambda\mapsto\widehat{\beta}_\lambda$ we get

Now, consider bootstrap samples.

for(s in 1:100){ id=sample(1:nrow(X),size=nrow(X),replace=TRUE) library(glmnet) vlambda=exp(seq(-4,2,length=201)) lasso=glmnet(x=X[id,],y=chicago[id,"Fire"],family="gaussian",alpha=1, lambda=vlambda,standardize=TRUE) plot(lasso,col=colrs,"lambda",lwd=.2,add=TRUE)}

We get here

The interpretation here is much more difficult

What about the order ?

N=matrix(NA,100000,4) for(s in 1:100000){ id=sample(1:nrow(X),size=nrow(X),replace=TRUE) library(glmnet) vlambda=exp(seq(-4,2,length=201)) lasso=glmnet(x=X[id,],y=chicago[id,"Fire"], family="gaussian",alpha=1, lambda=vlambda,standardize=TRUE) N[s,]=names(sort(apply(as.matrix(lasso$beta), 1,function(x) sum(x!=0))))} The ordering that was obtained on the original dataset was the same in 56% of the scenarios, mean(apply(N,1,function(x) paste(x,collapse="")=="(Intercept)X_1X_2X_3")) [1] 0.5693 We can look at all the cases, L=as.character(c(123,132,213,231,312,321)) Li=paste("(Intercept)X_",substr(L,1,1),"X_", substr(L,2,2),"X_",substr(L,3,3),sep="") g=function(y) mean(apply(N,1,function(x) paste(x,collapse="")==y)) vL=unlist(lapply(Li,g)) names(vL)=L barplot(vL,las=2,horiz=TRUE) # Standardization in LASSO The lasso regression is based on the idea of solving$$\widehat{\mathbf{\beta}}_{\lambda}=\text{argmin}\lbrace -\log\mathcal{L}(\mathbf{\beta}|\mathbf{x},\mathbf{y})+\lambda\|\mathbf{\beta}\|_{\ell_1}\rbrace$$where$$\Vert\mathbf{a} \Vert_{\ell_1}=\sum_{i=1}^d |a_i|$$for any $\mathbf{a}\in\mathbb{R}^d$. In a recent post, we’ve seen computational aspects of the optimization problem. But I went quickly throught the story of the $\ell_1$-norm. Because it means, somehow, that the value of $\beta_1$ and $\beta_2$ should be comparable. Somehow, with two significant variables, with very different scales, we should expect orders (or relative magnitudes) of $\widehat{\beta}_1$ and $\widehat{\beta}_2$ to be very very different. So people say that it is therefore necessary to center and reduce (or standardize) the variables. Consider the following (simulated) dataset Sigma = matrix(c(1,.8,.2,.8,1,.4,.2,.4,1),3,3) n = 1000 library(mnormt) X = rmnorm(n,rep(0,3),Sigma) set.seed(123) df = data.frame(X1=X[,1],X2=X[,2],X3=X[,3],X4=rnorm(n), X5=runif(n),X6=exp(X[,3]), X7=sample(c("A","B"),size=n,replace=TRUE,prob=c(.5,.5)), X8=sample(c("C","D"),size=n,replace=TRUE,prob=c(.5,.5))) df$Y = 1+df$X1-df$X4+5*(df$X7=="A")+rnorm(n) X = model.matrix(lm(Y~.,data=df)) Use the following colors for the graphs and the value of $\lambda$ library("RColorBrewer") colrs = c(brewer.pal(8,"Set1"))[c(1,4,5,2,6,3,7,8)] vlambda=exp(seq(-8,1,length=201)) The first regression we can run is a non-standardized one library(glmnet) lasso = glmnet(x=X,y=df[,"Y"],family="gaussian",alpha=1,lambda=vlambda,standardize=FALSE) We can visualize the graphs of $\lambda\mapsto\widehat{\beta}_\lambda$ idx = which(apply(lasso$beta,1,function(x) sum(x==0))&lt;200) plot(lasso,col=colrs,'lambda',xlim=c(-5.5,2.3),lwd=2) legend(1.2,.9,legend=paste('X',0:8,sep='')[idx],col=colrs,lty=1,lwd=2)

At least, observe that the most significant variables are the one that were used to generate the data.

Now, consider the case that we standardize the data

lasso = glmnet(x=X,y=df[,"Y"],family="gaussian",alpha=1,lambda=vlambda,standardize=TRUE)

The graphs of $\lambda\mapsto\widehat{\beta}_\lambda$

The graph is (strangely) very similar to the previous one. Except perhaps for the green curve. Maybe that categorical are not simular to continuous variables… Because somehow, standardisation of categorical variables might be not natural…

Why not consider some home-made function ? Let us transform (linearly) all variable in the $X$ matrix (except the first one, which is the intercept)

Xc = X for(j in 2:ncol(X)) Xc[,j]=(Xc[,j]-mean(Xc[,j]))/sd(Xc[,j])

Now, we can run our lasso regression on that one (with the intercept since all the variables are centered, but $y$)

lasso = glmnet(x=Xc,y=df$Y,family="gaussian",alpha=1,intercept=TRUE,lambda=vlambda) The plot is now plot(lasso,col=colrs,"lambda",xlim=c(-6.7,1.3),lwd=2) idx = which(apply(lasso$beta,1,function(x) sum(x==0))&lt;length(vlambda)) legend(.15,.45,legend=paste('X',0:8,sep='')[idx],col=colrs,lty=1,bty=&quot;n&quot;,lwd=2)

Actually, why not also center the $y$ variable, and remove also the intercept

Yc = (df[,"Y"]-mean(df[,"Y"]))/sd(df[,"Y"]) lasso = glmnet(x=Xc,y=Yc,family="gaussian",alpha=1,intercept=FALSE,lambda=vlambda)

Hopefully, those graphs are very consistent (and if we use those for variable selection, they suggest to use variables that were actually used to generate the dataset). And having qualitative and quantitative variable is not a big deal. But still, I do not feel confortable with the differences…

# Convex Regression Model

This morning during the lecture on nonlinear regression, I mentioned (very) briefly the case of convex regression. Since I forgot to mention the codes in R, I will publish them here. Assume that $y_i=m(\mathbf{x}_i)+\varepsilon_i$ where $m:\mathbb{R}^d\rightarrow \mathbb{R}$ is some convex function.

Then $m$ is convex if and only if $\forall\mathbf{x}_1,\mathbf{x}_2\in\mathbb{R}^d$, $\forall t\in[0,1]$, $$m(t\mathbf{x}_1+[1-t]\mathbf{x}_2) \leq tm(\mathbf{x}_1)+[1-t]m(\mathbf{x}_2)$$Hidreth (1954) proved that if$$m^\star=\underset{m \text{ convex}}{\text{argmin}}\left\lbrace\sum_{i=1}^n \big(y_i-m(\mathbf{x_i})\big)^2\right\rbrace$$then $\mathbf{\theta}^\star=(m^\star(\mathbf{x_1}),\cdots,m^\star(\mathbf{x_n}))$ is unique.

Let $\mathbf{y}=\mathbf{\theta}+\mathbf{\varepsilon}$, then $$\mathbf{\theta}^\star=\underset{\mathbf{\theta}\in \mathcal{K}}{\text{argmin}}\left\lbrace\sum_{i=1}^n \big(y_i-\theta_i)\big)^2\right\rbrace$$where$$\mathcal{K}=\{\mathbf{\theta}\in\mathbb{R}^n:\exists m\text{ convex },m(\mathbf{x}_i)=\theta_i\}$$. I.e. $\mathbf{\theta}^\star$ is the projection of $\mathbf{y}$ onto the (closed) convex cone $\mathcal{K}$. The projection theorem gives existence and unicity.

For convenience, in the application, we will consider the real-valued case, $m:\mathbb{R}\rightarrow \mathbb{R}$, i.e. $y_i=m(x_i)+\varepsilon_i$. Assume that observations are ordered $x_1\leq x_2\leq\cdots \leq x_n$. Here $$\mathcal{K}=\left\lbrace\mathbf{\theta}\in\mathbb{R}^n:\frac{\theta_2-\theta_1}{x_2-x_1}\leq \frac{\theta_3-\theta_2}{x_3-x_2}\leq \cdots \leq \frac{\theta_n-\theta_{n-1}}{x_n-x_{n-1}}\right\rbrace$$

Hence, quadratic program with $n-2$ linear constraints.

$m^\star$ is a piecewise linear function (interpolation of consecutive pairs $(x_i,\theta_i^\star)$).

If $m$ is differentiable, $m$ is convex if $$m(\mathbf{x})+ \nabla m(\mathbf{x})^{\text{T}}\cdot[\mathbf{y}-\mathbf{x}] \leq m(\mathbf{y})$$

More generally, if $m$ is convex, then there exists $\xi_{\mathbf{x}}\in\mathbb{R}^n$ such that $$m(\mathbf{x})+ \xi_{\mathbf{x}}^{\text{ T}}\cdot[\mathbf{y}-\mathbf{x}] \leq m(\mathbf{y})$$
$\xi_{\mathbf{x}}$ is a subgradient of $m$ at ${\mathbf{x}}$. And then $$\partial m(\mathbf{x})=\big\lbrace m(\mathbf{x})+ \xi^{\text{ T}}\cdot[\mathbf{y}-\mathbf{x}] \leq m(\mathbf{y}),\forall \mathbf{y}\in\mathbb{R}^n\big\rbrace$$

Hence, $\mathbf{\theta}^\star$ is solution of $$\text{argmin}\big\lbrace\|\mathbf{y}-\mathbf{\theta}\|^2\big\rbrace$$$$\text{subject to }\theta_i+\xi_i^{\text{ T}}[\mathbf{x}_j-\mathbf{x}_i]\leq\mathbf{\theta}_j,~\forall i,j$$ and $\xi_1,\cdots,\xi_n\in\mathbb{R}^n$. Now, to do it for real, use cobs package for constrained (b)splines regression,

library(cobs)

To get a convex regression, use

plot(cars) x = cars$speed y = cars$dist rc = conreg(x,y,convex=TRUE) lines(rc, col = 2)

Here we can get the values of the knots

rc   Call: conreg(x = x, y = y, convex = TRUE) Convex regression: From 19 separated x-values, using 5 inner knots, 7, 8, 9, 20, 23. RSS = 1356; R^2 = 0.8766; needed (5,0) iterations

and actually, if we use them in a linear-spline regression, we get the same output here

reg = lm(dist~bs(speed,degree=1,knots=c(4,7,8,9,,20,23,25)),data=cars) u = seq(4,25,by=.1) v = predict(reg,newdata=data.frame(speed=u)) lines(u,v,col="green")

Let us add vertical lines for the knots

abline(v=c(4,7,8,9,20,23,25),col="grey",lty=2)

# Classification from scratch, logistic with splines 2/8

Today, second post of our series on classification from scratch, following the brief introduction on the logistic regression.

## Piecewise linear splines

To illustrate what’s going on, let us start with a “simple” regression (with only one explanatory variable). The underlying idea is natura non facit saltus, for “nature does not make jumps”, i.e. process governing equations for natural things are continuous. That seems to be a rather strong assumption, because we can assume that there is a fixed threshold to explain death. For instance, if patients die (for sure) if the “stroke index” exceeds a threshold, we might expect some discontinuity. Exceept that if that threshold is an heterogeneous (non-observable continuous) variable, then we get back to the continuity assumption.

The most simple model we can think of to extend the linear model we’ve seen in the previous post is to consider a piecewise linear function, with two parts : small values of $x$, and larger values of $x$. The most convenient way to do so is to use the positive part function $(x-s)_+$ which is the difference between $x$ and $s$ if that difference is positive, and $0$ otherwise. For instance $$\beta_1 x+\beta_2(x-s)_+$$ is the following piecewise linear function, continuous, with a “rupture” at knot $s$.

Observe also the following interpretation: for small values of $x$, there is a linear increase, with slope $\beta_1$, and for lager values of $x$, there is a linear decrease, with slope $\beta_1+\beta_2$. Hence, $\beta_2$ is interpreted as a change of the slope.

And of course, it is possible to consider more than one knot. The function to get the positive value is the following

pos = function(x,s) (x-s)*(x&gt;=s)

then we can use it direcly in our regression model

reg = glm(PRONO~INSYS+pos(INSYS,15)+ pos(INSYS,25),data=myocarde,family=binomial)

The output of the regression is here

summary(reg)   Coefficients: Estimate Std. Error z value Pr(&gt;|z|) (Intercept) -0.1109 3.2783 -0.034 0.9730 INSYS -0.1751 0.2526 -0.693 0.4883 pos(INSYS, 15) 0.7900 0.3745 2.109 0.0349 * pos(INSYS, 25) -0.5797 0.2903 -1.997 0.0458 *

Hence, the original slope, for very small values is not significant, but then, above 15, it become significantly positive. And above 25, there is a significant change again. We can plot it to see what’s going on

u = seq(5,55,length=201) v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,type="l") points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=c(5,15,25,55),lty=2)

## Using bs() linear splines

Using the GAM function, things are slightly different. We will use here so called b-splines,

library(splines)

We can define spline functions with support $(5,55)$ and with knots $\{15,25\}$

clr6 = c("#1b9e77","#d95f02","#7570b3","#e7298a","#66a61e","#e6ab02") x = seq(0,60,by=.25) B = bs(x,knots=c(15,25),Boundary.knots=c(5,55),degre=1) matplot(x,B,type="l",lty=1,lwd=2,col=clr6)

as we can see, the functions defined here are different from the one before, but we still have (piecewise) linear functions on each segment $(5,15)$, $(15,25)$ and $(25,55)$. But linear combinations of those functions (the two sets of functions) will generate the same space. Said differently, if the interpretation of the output will be different, predictions should be the same

reg = glm(PRONO~bs(INSYS,knots=c(15,25), Boundary.knots=c(5,55),degre=1), data=myocarde,family=binomial) summary(reg)   Coefficients: Estimate Std. Error z value Pr(&gt;|z|) (Intercept) -0.9863 2.0555 -0.480 0.6314 bs(INSYS,..)1 -1.7507 2.5262 -0.693 0.4883 bs(INSYS,..)2 4.3989 2.0619 2.133 0.0329 * bs(INSYS,..)3 5.4572 5.4146 1.008 0.3135

Observe that there are three coefficients, as before, but again, the interpretation is here more complicated…

v=predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red") points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=c(5,15,25,55),lty=2)

Nevertheless, the prediction is the same… and that’s nice.

## Piecewise quadratic splines

Let us go one step further… Can we have also the continuity of the derivative ? Yes, and that’s easy actually, considering parabolic functions. Instead of using a decomposition on $x,(x-s_1)_+$ and $(x-s_2)_+$ consider now a decomposition on $x,x^{\color{red}{2}},(x-s_1)^{\color{red}{2}}_+$ and $(x-s_2)^{\color{red}{2}}_+$.

 pos2 = function(x,s) (x-s)^2*(x&gt;=s) reg = glm(PRONO~poly(INSYS,2)+pos2(INSYS,15)+pos2(INSYS,25), data=myocarde,family=binomial) summary(reg)   Coefficients: Estimate Std. Error z value Pr(&gt;|z|) (Intercept) 29.9842 15.2368 1.968 0.0491 * poly(INSYS, 2)1 408.7851 202.4194 2.019 0.0434 * poly(INSYS, 2)2 199.1628 101.5892 1.960 0.0499 * pos2(INSYS, 15) -0.2281 0.1264 -1.805 0.0712 . pos2(INSYS, 25) 0.0439 0.0805 0.545 0.5855

As expected, there are here five coefficients: the intercept and two for the part on the left (three parameters for the parabolic function), and then two additional terms for the part in the center – here $(15,25)$ – and for the part on the right. Of course, for each portion, there is only one degree of freedom since we have a parabolic function (three coefficients) but two constraints (continuity, and continuity of the first order derivative).

On a graph, we get the following

v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red",lwd=2,xlab="INSYS",ylab="") points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=c(5,15,25,55),lty=2)

## Using bs() quadratic splines

Of course, we can do the same with our R function. But as before, the basis of function is expressed here differently

 x = seq(0,60,by=.25) B=bs(x,knots=c(15,25),Boundary.knots=c(5,55),degre=2) matplot(x,B,type="l",xlab="INSYS",col=clr6)

If we run R code, we get

reg = glm(PRONO~bs(INSYS,knots=c(15,25), Boundary.knots=c(5,55),degre=2),data=myocarde, family=binomial) summary(reg)   Coefficients: Estimate Std. Error z value Pr(&gt;|z|) (Intercept) 7.186 5.261 1.366 0.1720 bs(INSYS, ..)1 -14.656 7.923 -1.850 0.0643 . bs(INSYS, ..)2 -5.692 4.638 -1.227 0.2198 bs(INSYS, ..)3 -2.454 8.780 -0.279 0.7799 bs(INSYS, ..)4 6.429 41.675 0.154 0.8774

But that’s not really a big deal since the prediction is exactly the same

v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red") points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=c(5,15,25,55),lty=2)

## Cubic splines

Last, but not least, we can reach the cubic splines. With our previous notions, we would consider a decomposition on (guess what) $x,x^2,x^{\color{red}{3}},(x-s_1)^{\color{red}{3}}_+,(x-s_2)^{\color{red}{3}}_+$, to get this time continuity, as well as continuity of the first two derivatives (and to get a very smooth function, since even variations will be smooth). If we use the bs function, the basis is the followin

B=bs(x,knots=c(15,25),Boundary.knots=c(5,55),degre=3) matplot(x,B,type="l",lwd=2,col=clr6,lty=1,ylim=c(-.2,1.2)) abline(v=c(5,15,25,55),lty=2)

and the prediction will now be

reg = glm(PRONO~bs(INSYS,knots=c(15,25), Boundary.knots=c(5,55),degre=3), data=myocarde,family=binomial) u = seq(5,55,length=201) v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red",lwd=2) points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=c(5,15,25,55),lty=2)

Two last things before concluding (for today), the location of the knots, and the extension to additive models.

## Location of knots

In many applications, we do not want to specify the location of the knots. We just want – say – three (intermediary) knots. This can be done using

reg = glm(PRONO~1+bs(INSYS,degree=1,df=4),data=myocarde,family=binomial)

We can actually get the locations of the knots by looking at

attr(reg$terms, "predvars")[[3]] bs(INSYS, degree = 1L, knots = c(15.8, 21.4, 27.15), Boundary.knots = c(8.7, 54), intercept = FALSE) which provides us with the location of the boundary knots (the minumun and the maximum from from our sample) but also the three intermediary knots. Observe that actually, those five values are just (empirical) quantiles quantile(myocarde$INSYS,(0:4)/4) 0% 25% 50% 75% 100% 8.70 15.80 21.40 27.15 54.00

If we plot the prediction, we get

v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red",lwd=2) points(myocarde$INSYS,myocarde$PRONO,pch=19) abline(v=quantile(myocarde$INSYS,(0:4)/4),lty=2) If we get back on what was computed before the logit transformation, we clealy see ruptures are the different quantiles B = bs(x,degree=1,df=4) B = cbind(1,B) y = B%*%coefficients(reg) plot(x,y,type="l",col="red",lwd=2) abline(v=quantile(myocarde$INSYS,(0:4)/4),lty=2)

Note that if we do specify anything about knots (number or location), we get no knots…

reg = glm(PRONO~1+bs(INSYS,degree=2),data=myocarde,family=binomial) attr(reg$terms, "predvars")[[3]] bs(INSYS, degree = 2L, knots = numeric(0), Boundary.knots = c(8.7,54), intercept = FALSE) and if we look at the prediction u = seq(5,55,length=201) v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red",lwd=2) points(myocarde$INSYS,myocarde$PRONO,pch=19) actually, it is the same as a quadratic regression (as expected actually) reg = glm(PRONO~1+poly(INSYS,degree=2),data=myocarde,family=binomial) v = predict(reg,newdata=data.frame(INSYS=u),type="response") plot(u,v,ylim=0:1,type="l",col="red",lwd=2) points(myocarde$INSYS,myocarde$PRONO,pch=19) ## Additive models Consider now the second dataset, with two variables. Consider here a model like $$\mathbb{P}[Y|X_1=x_1,X_2=x_2]=\frac{\exp[\eta(x_1,x_2)]}{1+\exp[\eta(x_1,x_2)]}$$ where $$\exp[\eta(x_1,x_2)]=\beta_0+\color{red}{s_1(x_1)}+\color{blue}{s_2(x_2)}$$ $$\color{red}{s_1(x_1)}=\beta_{1,0}x_1+\beta_{1,1}(x_1-s_{11})_++\beta_{1,2}(x_1-s_{12})_+$$ and $$\color{blue}{s_2(x_2)}=\beta_{2,0}x_2+\beta_{2,1}(x_2-s_{21})_++\beta_{2,2}(x_2-s_{22})_+$$ It might seem a little bit restrictive, but that’s actually the idea of additive models. reg = glm(y~bs(x1,degree=1,df=3)+bs(x2,degree=1,df=3),data=df,family=binomial(link = "logit")) u = seq(0,1,length=101) p = function(x,y) predict.glm(reg,newdata=data.frame(x1=x,x2=y),type="response") v = outer(u,u,p) image(u,u,v,xlab="Variable 1",ylab="Variable 2",col=clr10,breaks=(0:10)/10) points(df$x1,df$x2,pch=19,cex=1.5,col="white") points(df$x1,df$x2,pch=c(1,19)[1+(df$y=="1")],cex=1.5) contour(u,u,v,levels = .5,add=TRUE)

Now, if think about is, we’ve been able to get a “perfect” model, so, somehow, it seems no longer continuous…

persp(u,u,v,theta=20,phi=40,col="green"

Of course, it is… it is piecewise linear, with hyperplane, some being almost vertical.

And one can also consider piecewise quadratic functions

reg = glm(y~bs(x1,degree=2,df=3)+bs(x2,degree=2,df=3),data=df,family=binomial(link = "logit")) u = seq(0,1,length=101) p = function(x,y) predict.glm(reg,newdata=data.frame(x1=x,x2=y),type="response") v = outer(u,u,p) image(u,u,v,xlab="Variable 1",ylab="Variable 2",col=clr10,breaks=(0:10)/10) points(df$x1,df$x2,pch=19,cex=1.5,col="white") points(df$x1,df$x2,pch=c(1,19)[1+(df$y=="1")],cex=1.5) contour(u,u,v,levels = .5,add=TRUE) Funny thing, we now have two “perfect” models, with different areas for the white and the black dots… Don’t ask me how to choose on that one. In R, it is possible to use the mgcv package to run a gam regression. It is used for generalized additive models, but here, we have only one variable, so it is difficult to see the “additive” part, actually. And to be more specific, mgcv is using penalized quasi-likelihood from the nlme package (but we’ll get back on penalized routines later on). But maybe I should also mention another smoothing tool before, kernels (and maybe also $k$-nearest neighbors). To be continued # Classification from scratch, logistic regression 1/8 Let us start today our series on classification from scratch The logistic regression is based on the assumption that given covariates $\mathbf{x}$, $Y$ has a Bernoulli distribution,$$Y|\mathbf{X}=\mathbf{x}\sim\mathcal{B}(p_{\mathbf{x}}),~~~~p_\mathbf{x}=\frac{\exp[\mathbf{x}^T\mathbf{\beta}]}{1+\exp[\mathbf{x}^T\mathbf{\beta}]}$$The goal is to estimate parameter $\mathbf{\beta}$. Recall that the heuristics for the use of that function for the probability is that$$\log[\text{odds}(Y=1)]=\log\frac{\mathbb{P}[Y=1]}{\mathbb{P}[Y=0]}=\mathbf{x}^T\mathbf{\beta}$$ ## Maximimum of the (log)-likelihood function The log-likelihood is here$$\log\mathcal{L} = \sum_{i=1}^n y_i\log p_i+(1-y_i)\log (1-p_i)$$ where $p_{i}=(1+\exp[-\mathbf{x}_i^T\mathbf{\beta}])^{-1}$. Numerical techniques are based on (numerical) gradient descent to compute the maximum of the likelihood function. The (negative) log-likelihood is the following function y = myocarde$PRONO X = cbind(1,as.matrix(myocarde[,1:7])) negLogLik = function(beta){ -sum(-y*log(1 + exp(-(X%*%beta))) - (1-y)*log(1 + exp(X%*%beta))) }

We use the minus sign since standard optimization routines compute minima, not maxima. Now, to find the minimum of that function, we need a starting point to initiate the algorithm

beta_init = lm(PRONO~.,data=myocarde)$coefficients Why not start with the parameter of the OLS. Somehow, we might think that at least, sign should be ok for instance. Anyway, we need a starting point, and let us use that one. logistic_opt = optim(par = beta_init, negLogLik, hessian=TRUE, method = "BFGS", control=list(abstol=1e-9)) Here, we obtain  logistic_opt$par (Intercept) FRCAR INCAR INSYS 1.656926397 0.045234029 -2.119441743 0.204023835 PRDIA PAPUL PVENT REPUL -0.102420095 0.165823647 -0.081047525 -0.005992238

Let us verify here that this output is valid. For instance, what if we change the value of the starting point (randomly)

simu = function(i){ logistic_opt_i = optim(par = rnorm(8,0,3)*beta_init, negLogLik, hessian=TRUE, method = "BFGS", control=list(abstol=1e-9)) logistic_opt_i$par[2:3] } v_beta = t(Vectorize(simu)(1:1000)) plot(v_beta) par(mfrow=c(1,2)) hist(v_beta[,1],xlab=names(myocarde)[1]) hist(v_beta[,2],xlab=names(myocarde)[2]) Ooops. There is a problem here. Clearly, we cannot rely on numerical optimization here. We can think about using another optimization routine library(optimx) logit = function(mX, vBeta) { exp(mX %*% vBeta)/(1+ exp(mX %*% vBeta)) } logLikelihoodLogitStable = function(vBeta, mX, vY) { -sum(vY*(mX %*% vBeta - log(1+exp(mX %*% vBeta))) + (1-vY)*(-log(1 + exp(mX %*% vBeta)))) } likelihoodScore = function(vBeta, mX, vY) { return(t(mX) %*% (logit(mX, vBeta) - vY) ) } optimLogitLBFGS = optimx(beta_init, logLikelihoodLogitStable, method = 'L-BFGS-B', gr = likelihoodScore, mX = X, vY = y, hessian=TRUE) The optimum is here attr(optimLogitLBFGS, "details")[[2]] [,1] 0.066680272 FRCAR 0.003080542 INCAR 0.079031364 INSYS -0.001586194 PRDIA 0.040500697 PAPUL -0.041870705 PVENT -0.014162756 REPUL 0.195632244 Let’s be honest here, I do not feel confortable with those techniques. So, what happened here ? Here, the technique we use is based on the following idea,$$\mathbf{\beta}_{new}=\mathbf{\beta}_{old} -\left(\frac{\partial^2\log\mathcal{L}(\mathbf{\beta}_{old})}{\partial\mathbf{\beta}\partial\mathbf{\beta}^T}\right)^{-1}\cdot \frac{\partial\log\mathcal{L}(\mathbf{\beta}_{old})}{\partial\mathbf{\beta}}$$The problem is that my computer does not know this first and second derivatives. So it will compute them using approximation techniques. Actually, it is possible to use functions dedicated to such computation library(numDeriv) library(MASS) logit = function(x){1/(1+exp(-x))} logLik = function(beta, X, y){ -sum(y*log(logit(X%*%beta)) + (1-y)*log(1-logit(X%*%beta))) } optim_second = function(beta, num_iter){ LL = vector() for(i in 1:num_iter){ grad = (t(X)%*%(logit(X%*%beta) - y)) H = hessian(logLik, beta, method = "complex", X = X, y = y) beta = beta - ginv(H)%*%grad LL[i] = logLik(beta, X, y) } result = list(beta, H) return(result) } With our OLS starting point, we obtain opt0 = optim_second(beta_init,500) opt0[[1]] [,1] [1,] 0.951074420 [2,] 0.018860280 [3,] 0.275428978 [4,] 0.144803636 [5,] -0.058535606 [6,] 0.001182178 [7,] -0.108651776 [8,] -0.002940315 But if we try with another starting point opt1 = optim_second(beta_init*runif(8),500) opt1[[1]] [,1] [1,] 0.052894794 [2,] 0.024718435 [3,] 0.167953661 [4,] 0.171662947 [5,] -0.057458066 [6,] -0.011361034 [7,] -0.107532114 [8,] -0.002679064 Clearly, some coefficients are rather close. But other aren’t. From my point of viezw, that is a major problem (keep in mind that we do not deal here with massive data ! There are only 7 explanatory variables, and only 71 observations). Why not try to be clever, and use the analytical values of those derivatives ? Even if some people claim the oppositive, sometimes, it can actually be usefull to do the maths, instead of considering only numerical values. ## Newton (or Fisher) Algorithm If you open any Econometrics textbooks (one can also try to derive it), you will get $$\frac{\partial\log\mathcal{L}(\mathbf{\beta}_{old})}{\partial\mathbf{\beta}}=\mathbf{X}^T(\mathbf{y}-\mathbf{p}_{old})$$ while$$\frac{\partial^2\log\mathcal{L}(\mathbf{\beta}_{old})}{\partial\mathbf{\beta}\partial\mathbf{\beta}^T}=-\mathbf{X}^T\mathbf{\Delta}_{old}\mathbf{X}$$ Y=myocarde$PRONO X=cbind(1,as.matrix(myocarde[,1:7])) colnames(X)=c("Inter",names(myocarde[,1:7])) beta=as.matrix(lm(Y~0+X)$coefficients,ncol=1) for(s in 1:9){ pi=exp(X%*%beta[,s])/(1+exp(X%*%beta[,s])) gradient=t(X)%*%(Y-pi) omega=matrix(0,nrow(X),nrow(X));diag(omega)=(pi*(1-pi)) Hessian=-t(X)%*%omega%*%X beta=cbind(beta,beta[,s]-solve(Hessian)%*%gradient)} Observe that here, I use only ten iterations of the algorithm !  beta[,8:10] [,1] [,2] [,3] XInter -10.187641685 -10.187641696 -10.187641696 XFRCAR 0.138178119 0.138178119 0.138178119 XINCAR -5.862429035 -5.862429037 -5.862429037 XINSYS 0.717084018 0.717084018 0.717084018 XPRDIA -0.073668171 -0.073668171 -0.073668171 XPAPUL 0.016756506 0.016756506 0.016756506 XPVENT -0.106776012 -0.106776012 -0.106776012 XREPUL -0.003154187 -0.003154187 -0.003154187 The thing is that is seems to converge extremely fast. And it is rather robust ! Look at what we get if we change our starting point beta=as.matrix(lm(Y~0+X)$coefficients,ncol=1)*runif(8) for(s in 1:9){ pi=exp(X%*%beta[,s])/(1+exp(X%*%beta[,s])) gradient=t(X)%*%(Y-pi) omega=matrix(0,nrow(X),nrow(X));diag(omega)=(pi*(1-pi)) Hessian=-t(X)%*%omega%*%X beta=cbind(beta,beta[,s]-solve(Hessian)%*%gradient)} beta[,8:10] [,1] [,2] [,3] XInter -10.187641586 -10.187641696 -10.187641696 XFRCAR 0.138178118 0.138178119 0.138178119 XINCAR -5.862429017 -5.862429037 -5.862429037 XINSYS 0.717084013 0.717084018 0.717084018 XPRDIA -0.073668172 -0.073668171 -0.073668171 XPAPUL 0.016756508 0.016756506 0.016756506 XPVENT -0.106776012 -0.106776012 -0.106776012 XREPUL -0.003154187 -0.003154187 -0.003154187

Nice, isn’t it? Looks like we got our winner, don’t we? And one can use the inverse of the Hessian matrix to get standard deviations.

## Weighted Least-Squares

Let us go one step further. We’ve seen that we want to compute something like$$\mathbf{\beta}_{new} =(\mathbf{X}^T\mathbf{\Delta}_{old}\mathbf{X})^{-1}\mathbf{X}^T\mathbf{\Delta}_{old}\mathbf{z}$$(if we do substitute matrices in the analytical expressions) where $\mathbf{z}=\mathbf{X}\mathbf{\beta}_{old}+\mathbf{\Delta}_{old}^{-1}[\mathbf{y}-\mathbf{p}_{old}]$. But actually, that’s simply a standard least-square problem$$\mathbf{\beta}_{new} = \text{argmin}\left\lbrace(\mathbf{z}-\mathbf{X}\mathbf{\beta})^T\mathbf{\Delta}_{old}^{-1}(\mathbf{z}-\mathbf{X}\mathbf{\beta})\right\rbrace$$The only problem here is that weights $\mathbf{\Delta}_{old}$ are functions of unknown $\mathbf{\beta}_{old}$. But actually, if we keep iterating, we should be able to solve it : given the $\mathbf{\beta}$ we got the weights, and with the weights, we can use weighted OLS to get an updated $\mathbf{\beta}$. That’s the idea of iteratively reweighted least squares.

The algorithm will be

df = myocarde beta_init = lm(PRONO~.,data=df)$coefficients X = cbind(1,as.matrix(myocarde[,1:7])) beta = beta_init for(s in 1:1000){ p = exp(X %*% beta) / (1+exp(X %*% beta)) omega = diag(nrow(df)) diag(omega) = (p*(1-p)) df$Z = X %*% beta + solve(omega) %*% (df$PRONO - p) beta = lm(Z~.,data=df[,-8], weights=diag(omega))$coefficients }

and the output is here

 beta (Intercept) FRCAR INCAR INSYS PRDIA -10.187641696 0.138178119 -5.862429037 0.717084018 -0.073668171 PAPUL PVENT REPUL 0.016756506 -0.106776012 -0.003154187

which is almost what we’ve obtained before. Nice isn’t it ? Actually, here we also have standard deviations of estimators

summary( lm(Z~.,data=df[,-8], weights=diag(omega)))   Coefficients: Estimate Std. Error t value Pr(&gt;|t|) (Intercept) -10.187642 10.668138 -0.955 0.343 FRCAR 0.138178 0.102340 1.350 0.182 INCAR -5.862429 6.052560 -0.969 0.336 INSYS 0.717084 0.503527 1.424 0.159 PRDIA -0.073668 0.261549 -0.282 0.779 PAPUL 0.016757 0.306666 0.055 0.957 PVENT -0.106776 0.099145 -1.077 0.286 REPUL -0.003154 0.004386 -0.719 0.475

## The standard glm function

Of course, it is possible to use an R built-in function to get our estimate

summary(glm(PRONO~.,data=myocarde,family=binomial(link = "logit")))   Coefficients: Estimate Std. Error z value Pr(&gt;|z|) (Intercept) -10.187642 11.895227 -0.856 0.392 FRCAR 0.138178 0.114112 1.211 0.226 INCAR -5.862429 6.748785 -0.869 0.385 INSYS 0.717084 0.561445 1.277 0.202 PRDIA -0.073668 0.291636 -0.253 0.801 PAPUL 0.016757 0.341942 0.049 0.961 PVENT -0.106776 0.110550 -0.966 0.334 REPUL -0.003154 0.004891 -0.645 0.519

## Application and visualisation

Let us visualize the prediction obtained from the logistic regression, on our second dataset

x = c(.4,.55,.65,.9,.1,.35,.5,.15,.2,.85) y = c(.85,.95,.8,.87,.5,.55,.5,.2,.1,.3) z = c(1,1,1,1,1,0,0,1,0,0) df = data.frame(x1=x,x2=y,y=as.factor(z)) reg = glm(y~x1+x2,data=df,family=binomial(link = "logit")) u = seq(0,1,length=101) p = function(x,y) predict.glm(reg,newdata=data.frame(x1=x,x2=y),type="response") v = outer(u,u,p) image(u,u,v,xlab="Variable 1",ylab="Variable 2",col=clr10,breaks=(0:10)/10) points(x,y,pch=19,cex=1.5,col="white") points(x,y,pch=c(1,19)[1+z],cex=1.5) contour(u,u,v,levels = .5,add=TRUE)

Here level curves – or iso-probabilities – are linear, so the space is divided in two (0 and 1, survival and death, white and black) by a straight line (or an hyperplane in higher dimension). Furthermore, since we have a linear model, if we change the cutoff (the threshold used to create the two classes), we obtain another straight line (or hyperplane) parallel to the first one.

Next time, we will introduce splines to smooth those continuous covariates… to be continued.

# Classification from scratch, overview 0/8

Before my course on « big data and economics » at the university of Barcelona in July, I wanted to upload a series of posts on classification techniques, to get an insight on machine learning tools.

According to some common idea, machine learning algorithms are black boxes. I wanted to get back on that saying. First of all, isn’t it the case also for regression models, like generalized additive models (with splines) ? Do you really know what the algorithm is doing ? Even the logistic regression. In textbooks, we can easily find math formulas. But what is really done when I run it, in R ?

When I started working on academia, someone told me something like « if you really want to understand a theory, teach it ». And that has been my moto for more than 15 years. I wanted to add a second part to that statement: « if you really want to understand an algorithm, recode it ». So let’s try this… My ambition is to recode (more or less) most of the standard algorithms used in predictive modeling, from scratch, in R. What I plan to mention, within the next two weeks, will be

I will use two datasets to illustrate. The first one is inspired by the cover of « Foundations of Machine Learning » by Mehryar Mohri, Afshin Rostamizadeh and Ameet Talwalkar. At least, with this dataset, it will be possible to plot predictions (since there are only two – continuous – features)

x = c(.4,.55,.65,.9,.1,.35,.5,.15,.2,.85) y = c(.85,.95,.8,.87,.5,.55,.5,.2,.1,.3) z = c(1,1,1,1,1,0,0,1,0,0) df = data.frame(x1=x,x2=y,y=as.factor(z)) plot(x,y,pch=c(1,19)[1+z])

Here is some code to get a visualization of the prediction (here the probability to be a black point)

rmatrix_model = function(model){ u = seq(0,1,length=101) p = function(x,y) predict(model,newdata=data.frame(x1=x,x2=y),type="response") v = outer(u,u,p) return(v)} nice_graph=function(v){ u = seq(0,1,length=101) image(u,u,v,xlab="Variable 1",ylab="Variable 2",col=clr10[c(1,10)],breaks=c(0,5,10)/10) points(x,y,pch=19,cex=1.5,col="white") points(x,y,pch=c(1,19)[1+z],cex=1.5) contour(u,u,v,levels = .5,add=TRUE) } reg = glm(y~x1+x2,data=df,family=binomial) nice_graph(rmatrix_model(reg))

Note that colors are defined here as

clr10= c("#ffffff","#f7fcfd","#e5f5f9","#ccece6","#99d8c9","#66c2a4","#41ae76","#238b45","#006d2c","#00441b")

or with some nonlinear model

The second one is a dataset I got from Gilbert Saporta, about heart attacks and decease (our binary variable).

myocarde = read.table("http://freakonometrics.free.fr/myocarde.csv",head=TRUE, sep=";") myocarde$PRONO = (myocarde$PRONO=="SURVIE")*1 y = myocarde\$PRONO X = as.matrix(cbind(1,myocarde[,1:7]))

So far, I do not plan to talk (too much) on the choice of tunning parameters (and cross-validation), on comparing models, etc. The goal here is simply to understand what’s going on when we call either glm, glmnet, gam, random forest, svm, xgboost, or any function to get a predict model.