More importantly than the statement from Sturla, which I may or may not
agree with based on the modeling assumption (and every p-value is based
on a modeling assumption), the logistic in scikit-learn is a penalized
logistic model. Thus the closed-form formulas for p-values are not valid.
G
On Sa
On Sun, Apr 19, 2015 at 9:26 AM, Alan G Isaac wrote:
> It seems unlikely that the choice of which features to provide
> should turn entirely on controversial philosophical positions.
> Hopefully a feature can be declared in or out of scope for the
> project on technical grounds.
But some design d
It seems unlikely that the choice of which features to provide
should turn entirely on controversial philosophical positions.
Hopefully a feature can be declared in or out of scope for the
project on technical grounds.
Alan Isaac
On 4/18/2015 7:20 PM, josef.p...@gmail.com wrote:
> Sturla means: No
It wouldn't hurt to have p-values returned, but personally, I don't miss them
in scikit-learn. I think that's a classic "ML vs. statistics" discussion --
what I mean is the inference vs. prediction stuff. To me, scikit-learn is
primarily a machine learning library.
> On Apr 19, 2015, at 12:53 A
wrote:
> Good, I was reading your previous comments on the topic as being
> against all frequentist null hypothesis testing.
In the frequentist paradigm I prefer to use model selection instead of
classical hypothesis testing with p-values. My focus is on building useful
models which are able to
wrote:
> Note. The editors of Basic and Applied Social Psychology are also
> banning confidence intervals.
I know. I am not sure I agree on that. I don't particularly like confidence
intervals very much, but I don't hate them with a passion.
Pro: Even though confidence intervals have a bizarre
On Sat, Apr 18, 2015 at 9:45 PM, Sturla Molden wrote:
> wrote:
>
>> (I just went through some articles to see how we can produce p-values
>> after feature selection with penalized least squares or maximum
>> penalized likelihood. :)
>
> If you have used penalized least squares or penalized likeli
On Sat, Apr 18, 2015 at 9:25 PM, Sturla Molden wrote:
> wrote:
>
>>> Re. "We should therefore never compute p-values": I assume that you meant
>>> that within the narrow context of regression, and not, e.g., in the context
>>> of tests of distribution.
>>
>> Sturla means: No null hypothesis testi
wrote:
> (I just went through some articles to see how we can produce p-values
> after feature selection with penalized least squares or maximum
> penalized likelihood. :)
If you have used penalized least squares or penalized likelihood, you have
already pruned the model for parameters that only
wrote:
>> Re. "We should therefore never compute p-values": I assume that you meant
>> that within the narrow context of regression, and not, e.g., in the context
>> of tests of distribution.
>
> Sturla means: No null hypothesis testing at all
Not really, I mean "no p-values for inferential sta
On Sat, Apr 18, 2015 at 6:40 PM, Phillip Feldman
wrote:
> This is a very nice explanation. Thanks!!
>
> Re. "We should therefore never compute p-values": I assume that you meant
> that within the narrow context of regression, and not, e.g., in the context
> of tests of distribution.
Sturla means
This is a very nice explanation. Thanks!!
Re. "We should therefore never compute p-values": I assume that you meant
that within the narrow context of regression, and not, e.g., in the context
of tests of distribution.
On Sat, Apr 18, 2015 at 3:31 PM, Sturla Molden
wrote:
> Phillip Feldman
> w
Phillip Feldman
wrote:
> When using logistic regression, I'm often trying to establish whether a
> given feature has any effect.
Compare models with and without the feature: Cross-validation, BIC, AIC,
PRESS, Bayes factor, etc. By the rules of inductive reasoning (cf. lex
parsimoniae, Occam's
I was able to accomplish what I needed using
`statsmodels.discrete.discrete_model.Logit`. Thanks!
On Sat, Apr 18, 2015 at 11:35 AM, Michael Kneier
wrote:
> Hi Phillip,
>
> Have you checked out statsmodel? That might be a better fit for your needs.
>
> Sent from my iPhone
>
> > On Apr 18, 2015,
Hi Phillip,
Have you checked out statsmodel? That might be a better fit for your needs.
Sent from my iPhone
> On Apr 18, 2015, at 8:31 PM, Phillip Feldman
> wrote:
>
> When using logistic regression, I'm often trying to establish whether a given
> feature has any effect. R and Matlab give m
When using logistic regression, I'm often trying to establish whether a
given feature has any effect. R and Matlab give me p-values, but
Scikit-learn does not. I would love to be able to do all of my statistical
processing in Python. Please consider adding this feature.
Phillip M. Feldman
-
16 matches
Mail list logo
