Why is SGDClassifier not listening to configuration files (mlspl.conf)?

[Kermit1007]

Hi,

My classifier (SGDClassifier) is allowing only 100 distinct cat values. I followed the link Configure algorithm performance costs - Splunk Documentation and modified the file mlspl.conf as follows :

[SGDClassifier]
max_distinct_cat_values=2000
max_distinct_cat_values_for_classifiers=2000

However my splunk search is giving me the error :

I had already change this value for another classifier (LinearSVC) and everything was fine, what did i miss here ? I just copied and pasted from it and change the Stanza name 

I'm using MLTK 5.3.1

[Kermit1007 · Accepted solution]

Update :

I have found the culprit :

In apps/Splunk_ML_Toolkit/bin/algos/SGDClassifier.py

at line 78 AND 116 

Change line 

to

 (There is a missing 'S' in the config parameter)

I think this is a bug from splunk and woud like to report it to the MLTK Team do you know a way ?