mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dmitriy Lyubimov <dlie...@gmail.com>
Subject Re: Regarding find p-value in logistic regression
Date Thu, 07 Feb 2013 20:53:53 GMT
I had same question once, i wonder to hear the answer to this one again.

As far as i understand, in R this value is derived analytically whereas
Mahout's regression is stochastic and as far as i understand those
techniques are not conducive to direct application of computation done in
R. But i wonder if such approximate inference of the p-value on the
regressor coefficient could be applied . (Not that i totally understand
analysis of variance done in R either for these guys).


On Thu, Feb 7, 2013 at 3:26 AM, prabu palanisamy <prabu@serendio.com> wrote:

> Hi
>
> In "R" Language, the Logistic Regression function "glm"  generates the "P
> Value".
>
> In Mahout's "TrainLogistic" - the logistic regression function provides
> only the coefficients of the variable .
>
> Is there anyway to get the P value in Mahout as  "R's glm" provides P
> Value.
>
> *R's glm Example Input*
> ** mylogit <- glm(IsAlert ~ P1 + P2, data=mysample, family="binomial")
>  summary(mylogit)
>  *
> **R's glm Output*
>
> Coefficients:
>                    Estimate     Std. Error      z value       Pr(>|z|) (*P
> values*)
> (Intercept)    9.597e-01    1.703e-01      5.634         1.76e-08 ***
> P1              -1.531e-02    7.232e-04     -21.167       < 2e-16 ***
> P2               7.353e-04    1.390e-03      0.529         0.597
>
> *Mahout, TrainLogistic Example Input:
>
> *$MAHOUT_HOME/bin/mahout trainlogistic --input mysample.csv \
> --output ./model \
> --target y --categories 2 \
> --predictors P1 P2 --types numeric \
> --features 2 --passes 100 --rate 50
>
> *Mahout OUTPUT*
> y ~ -0.157*Intercept Term + -0.678*P1 + -0.416*P2
> Intercept Term -0.15655
> P1 -0.67841
> P2 -0.41587.
>
> But I need to get P Value,
>
> Thanks in advance
>
> Thanks and Regards
> Prabu
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message