Avoid computing p-values in logistic regression

Highlighted
mouch
Level 1
Avoid computing p-values in logistic regression
Jump to solution

Hi, 



I am training a logistic regression in the doctor, however after training and scoring of the model (according to the logs) I have a memory error. I think this error is due to the computation of features to explain the model, is it possible to chose not to compute these features ?




[2018-05-29 15:15:21,674] [13200/MainThread] [INFO] [root] Output params are {'resolved': {'logit': {'penalty': 'l2', 'multi_class': u'ovr', 'C': 0.01}, 'algorithm': u'LOGISTIC_REGRESSION'}, 'other': {}}
[2018-05-29 15:15:21,674] [13200/MainThread] [INFO] [root] END - Fitting model
[2018-05-29 15:15:21,674] [13200/MainThread] [INFO] [root] START - Saving model
[2018-05-29 15:15:21,679] [13200/MainThread] [INFO] [root] PICKLING LogisticRegression(C=0.01, class_weight=None, dual=False, fit_intercept=True,
intercept_scaling=1, max_iter=100, multi_class=u'ovr', n_jobs=1,
penalty='l2', random_state=1337, solver='liblinear', tol=0.0001,
verbose=0, warm_start=False)
[2018-05-29 15:15:21,681] [13200/MainThread] [INFO] [root] END - Saving model
[2018-05-29 15:15:21,681] [13200/MainThread] [INFO] [root] START - Scoring model
[2018-05-29 15:15:21,682] [13200/MainThread] [INFO] [root] Intrinsic scoring
[2018-05-29 15:15:21,682] [13200/MainThread] [INFO] [root] Computing regression coeffs
[2018-05-29 15:15:21,682] [13200/MainThread] [INFO] [root] FEATURES 30 CLF COEF 30
[2018-05-29 15:15:21,682] [13200/MainThread] [INFO] [root] CLF Intercept: [-1.87283529]
[2018-05-29 15:15:21,812] [13200/MainThread] [INFO] [root] Coefs: [ -1.87283529e+00 -2.74150938e-01]
[2018/05/29-15:16:33.161] [FT-TrainWorkThread-G7UFDiiP-4863] [INFO] [dku.analysis.ml.python] T-4dKTnPjI - Processing thread joined ...
[2018/05/29-15:16:33.163] [FT-TrainWorkThread-G7UFDiiP-4863] [INFO] [dku.analysis.ml.python] T-4dKTnPjI - Joining processing thread ...
[2018/05/29-15:18:05.229] [KNL-python-single-command-kernel-monitor-4872] [INFO] [dku.kernels] - Process done with code 137
[2018/05/29-15:18:05.246] [KNL-python-single-command-kernel-monitor-4872] [INFO] [dip.tickets] - Destroying API ticket for analysis-ml-***-nnzrg12 on behalf of admin
[2018/05/29-15:18:05.247] [MRT-4865] [INFO] [dku.kernels] - Getting kernel tail
[2018/05/29-15:18:05.352] [MRT-4865] [INFO] [dku.kernels] - Getting kernel tail
[2018/05/29-15:18:05.354] [MRT-4865] [ERROR] [dku.analysis.prediction] - Processing failed
com.dataiku.dip.exceptions.ProcessDiedException: Process died (exit code: 137, killed - maybe out of memory ?)


Thanks for the help!



Max

0 Kudos
1 Solution

Accepted Solutions
Alex_Combessie Dataiker
Dataiker
Re: Avoid computing p-values in logistic regression
Jump to solution
Hey Mouch Mouch,

Good to see you. Updating DSS to the latest version should solve the issue!

Cheers,

Alex

View solution in original post

0 Kudos
3 Replies
Alex_Combessie Dataiker
Dataiker
Re: Avoid computing p-values in logistic regression
Jump to solution
Hey Mouch Mouch,

Good to see you. Updating DSS to the latest version should solve the issue!

Cheers,

Alex

View solution in original post

0 Kudos
mouch
Level 1
Re: Avoid computing p-values in logistic regression
Jump to solution
Thanks Alex!
It worked like a charm!

Max
0 Kudos
Alex_Combessie Dataiker
Dataiker
Re: Avoid computing p-values in logistic regression
Jump to solution
thank Clément 🙂
0 Kudos
Labels (1)