Je préfère également dire à mes étudiants qu’on «moissonne» des données:

https://jhroy.gitbooks.io/edm5240-h2017/content/semaine-07—python-3.html ]]>

Très intéréssent, j’essaye de regarder pour les sports que je pratique.

Deux soucis pour l’instant :

– Une typo, Url_Licenses devient Url_licences dans la ligne suivante

– L’URL http://freakonometrics.free.fr/Projet_R/Code_federation.csv est en erreur 404 ]]>

Its necessary run in a train and test dataset to get variable importance with Random forest?

]]>For 1) the ML system should sample a=1…n scenarios with x= Variance of cyclists starting times as predictor feature regressed over y= Variance of Temperature at x (as well as x+1 ; x-1 …) as target feature. As this is a tensor of brutal force model comparisons it would fit in well with ML applications and positively would solve for directionality issue of causality (I.e. It is x preceeding y or vice versa). In this sense ‘research designs’ would apply panel methods to solve for directionality of causal models.

For 2) and that’s interesting indeed as you might work out some data analitics on these data: “single variable” causality is a unicorn 🦄…..The coexistence of billion…infinite interactive covariates as !causal complex! is what Nature is teaching since we were monkeys….so I suggest here a stupid approximation by ANCOVA in R

DV=N of daily cyclisists

Factor=daily temperature

[level 1=-20:-10 ; level 2=-10:-5 ; level 3=-5:0 …etc]

Covariates=traffic, age, gender, seasonality, week day etc…

## the results should uncover the actual effect of number of ciclista 🚴 Reducing (covarying out of the model) the effect of several other key features …

Note: Partial (or semipartial) correlation models apply as well in this case to understand what mediator variables are present “between” cyclists and temperature.

For 3)

x -> y [for any x ‘increase temperature’ it is always an increase of y ‘🚵’ in Helsinki]

is logically true if

¬y -> ¬x [which is clearly false…]

Overall 1), 2) and 3) are implementable in a ML application to generalize (always tentatively!) causal order solutions by numerical/logical approximation and keeping the ML system as an agnostic model.

PS the real info hidden in your graph is the “heteroscedasticity”. Suffice to say that for non linear models like the one picked up by your graph a profit or logic function would give abnormal standard errors with biased MLE … also heteroscedasticity would be revealing presence of non-homogeneous sub-populations in the sample and thus ….other features would most likely be necessary to disentangle causality of the model .

]]>I took the liberty of reproducing your results (in “tidyverse” code”) and added some notes on efficiency. I have an R notebook/markdown here: https://github.com/ytoren/reproducible/tree/master/linear-regression-map-reduce

Keep up the good work mate!

]]>I think you might have a typo :

You split Q2 to Q2list by:

> for(j in 1:m) Q2list[[j]] = Q1[(j-1)*2+1:2,]

And it should be

> for(j in 1:m) Q2list[[j]] = Q2[(j-1)*2+1:2,]

P.S., On your other post you use the bootstrap with LASSO, which provides a great visualization. Though, I have heard Rob Tibshirani mention that BS would not be appropriate for getting precision values if that was ever the purpose of using it.

Thanks again,

Hayden

]]>It’s a truly staggering sad Fact.

I believe the longest period of peace was under FDR 1936-1941.

War is financed, therefore people make profit, like the Carlyle group.

The Bush family are a big part of this.

Prescott Bush was a shady figure.

I really don’t think America can stop now, why would it ? It’s a huge money machine.

After the Russian Revolution of 1917, the ultra rich became scared.

This is NOT a conspiracy theory, it’s a Fact.

It’s their World, we are just living in it. ]]>

I think you should use $ {\ell}_{n} $ for Infinite Sequences.

You should use $ {L}_{n} $ for Finite Spaces.

In you post you use $ {\ell}_{n} $ for finite dimension vector.

]]>You use $ {\ell}_{1} $ which was my first thought.

Yet I was told to be mistaken on Math Exchange and I should use $ {L}_{1} $ as $ {\ell}_{n} $ is reserved for infinite series which are summable with respect to the $ {\ell}_{n} $ norm.