Weekly Report - 18/11/2016

[amritbhanu]

DOING:

• Working on these todo lists #6 :

  Running:

• Does stability help classification? Tuned and untuned? Multigoal fscore and jaccard score
• Heat Map with tags and topics for different datasets. (With tuning)
• Tune svm and lda, just for fscore (kernel, C, degree, learning rate)

  roadblocks:

• Nothing yet

  admin:

• No

[timm]

ok

[timm]

we need those charts on "tuning improves classification" as part of the icse replies

[amritbhanu]

on it

[amritbhanu]

Results:

• Solid lines represents tuned results and dotted lines untuned.
• tuning improves f2scores by minimum 25-30% than untuned settings.
• tuning improves raw scores (jaccard) by minimum 80% and maximum of 200%