some stats

[as1986]

Predictions for both the fine grained labels, and coarse grained labels - in following experimental conditions-

1. training from all combined:
   1. predictions on all combined
      • 'python getAnswer.py 20 --accuracy > ../annotations/catAll.libsvm.wopron.accuracy'
      • 'python getAnswer.py 20 --accuracy > ../annotations/catAll.libsvm.wopron.coarse.accuracy'
2. training from 4Teds separately combined:
   1. predictions for 4TEDs
      • 'python getAnswer.py 20 --accuracy > ../annotations/catTed.libsvm.wopron.accuracy'
      • 'python getAnswer.py 20 --accuracy > ../annotations/catTed.libsvm.wopron.coarse.accuracy'
   2. predictions on episode (unseen data, different genre)
      • 'python getAnswer.py 1 --accuracy > ../annotations/catStoryOnCatTed.libsvm.wopron.accuracy'
      • 'python getAnswer.py 1 --accuracy > ../annotations/catStoryOnCatTed.libsvm.wopron.coarse.accuracy'
3. training from episode (probably we already know this will be bad):
   1. predictions for 4TEDs (unseen data, different genre)
      • 'python getAnswer.py 1 --accuracy > ../annotations/catTedOnCatStory.libsvm.wopron.accuracy'
      • 'python getAnswer.py 1 --accuracy > ../annotations/catTedOnCatStory.libsvm.wopron.coarse.accuracy'
   2. predictions for episode
      • 'python getAnswer.py 20 --accuracy > ../annotations/catStory.libsvm.wopron.accuracy'
      • 'python getAnswer.py 20 --accuracy > ../annotations/catStory.libsvm.wopron.coarse.accuracy'
4. training from 3TEDs (leave Alisa news TED out) + episode
   1. predictions for all of them
      • 'python getAnswer.py 20 --accuracy > ../annotations/catAll.libsvm.wopron.woalisa.accuracy'
      • 'python getAnswer.py 20 --accuracy > ../annotations/catAll.libsvm.wopron.woalisa.coarse.accuracy'
   2. predictions for Alisa news TED (unseen data, trng from combined genre)
      • 'python getAnswer.py 1 --accuracy > ../annotations/alisaOnCatAll.libsvm.wopron.accuracy'
      • 'python getAnswer.py 1 --accuracy > ../annotations/alisaOnCatAll.libsvm.wopron.coarse.accuracy'
5. training from 3TEDs (leave Alisa news TED out) , also leave episode out
   1. predictions for all of them
      • 'python getAnswer.py 20 --accuracy > ../annotations/catTed.libsvm.wopron.woalisa.accuracy'
      • 'python getAnswer.py 20 --accuracy > ../annotations/catTed.libsvm.wopron.woalisa.coarse.accuracy'
   2. predictions for Alisa news TED (unseen data, trng from same genre)
      • 'python getAnswer.py 1 --accuracy > ../annotations/alisaOnCatTed.libsvm.wopron.accuracy'
      • 'python getAnswer.py 1 --accuracy > ../annotations/alisaOnCatTed.libsvm.wopron.coarse.accuracy'