PUBDEV-8947: aic glm #6672
PUBDEV-8947: aic glm #6672
Conversation
…oving duplicate calculation
…un - removing duplicate calculation" This reverts commit f9627e8.
…rom GLMParameters
|
@syzonyuliia : I have added a R unit test for you that attempts to compare likelihood/aic calculation for GLM. You can do a |
|
Something is wrong. Here when I run the comparison with R, here is what I got:
|
…rom GLMParameters
syzonyuliia
changed the base branch from
master
to
HEXDEV-785-loglikelihood-GLM
syzonyuliia
changed the base branch from
HEXDEV-785-loglikelihood-GLM
to
master
| R.formula = (R.data[,"GLEASON"]~.) | ||
| model.h2o.gaussian.identity <- h2o.glm(calc_like=TRUE, x=myX, y=myY, training_frame=h2o.data, family="gaussian", link="identity",alpha=0.5, lambda=0, nfolds=0) | ||
| model.R.gaussian.identity <- glm(formula=R.formula, data=R.data[,2:9], family=gaussian(link=identity), na.action=na.omit) | ||
| perf <- h2o.performance(model.h2o.gaussian.identity, h2o.data) |
There was a problem hiding this comment.
change
perf <- h2o.performance(model.h2o.gaussian.identy, h2o.data)
to
perf <- h2o.performance(model.h2o.gaussian.identy, train=TRUE)
h2o-r/tests/testdir_algos/glm/runit_PUBDEV_8947_GLM_likelihoods_aic.R
Outdated
Show resolved
Hide resolved
h2o-r/tests/testdir_algos/glm/runit_PUBDEV_8947_GLM_likelihoods_aic.R
Outdated
Show resolved
Hide resolved
h2o-r/tests/testdir_algos/glm/runit_PUBDEV_8947_GLM_likelihoods_aic.R
Outdated
Show resolved
Hide resolved
…ative_binomial' into PUBDEV-8947-AIC-GLM
… dispersion parameter
| - log(yr) - Gamma.logGamma(w * invPhiEst); | ||
| case tweedie: | ||
| if (DispersionMethod.ml.equals(_dispersion_parameter_method) && !_fix_tweedie_variance_power) { | ||
| return -TweedieVariancePowerMLEstimator.logLikelihood(yr, ym[0], _tweedie_variance_power, _init_dispersion_parameter); |
There was a problem hiding this comment.
Change
if (DispersionMethod.ml.equals(_dispersion_parameter_method) && !_fix_tweedie_variance_power)
to just
if (DispersionMethod.ml.equals(_dispersion_parameter_method))
We should not care what the user sets when they want to estimate Tweedie dispersion parameter. They should be able to get their likelihood calculation regardless.
There was a problem hiding this comment.
Oh right, and we also explicitly set the _dispersion_parameter_method to ML, so there is no reason to check for it
https://h2oai.atlassian.net/browse/PUBDEV-8947