facebookresearch/StarSpace

Duplicate Results When Predict for Classification

ZhaoyangChen opened this issue · 2 comments

ZhaoyangChen commented

According to demo

 ./starspace train \
  -trainFile "${DATADIR}"/ag_news.train \
  -model "${MODELDIR}"/ag_news \
  -initRandSd 0.01 \
  -adagrad false \
  -ngrams 1 \
  -lr 0.01 \
  -epoch 5 \
  -thread 20 \
  -dim 10 \
  -negSearchLimit 5 \
  -trainMode 0 \
  -label "__label__" \
  -similarity "dot" \
  -verbose true

echo "Start to evaluate trained model:"

./starspace test \
  -model "${MODELDIR}"/ag_news \
  -testFile "${DATADIR}"/ag_news.test \
  -ngrams 1 \
  -dim 10 \
  -label "__label__" \
  -thread 10 \
  -similarity "dot" \
  -trainMode 0 \
  -verbose true \
  -predictionFile '/tmp/starspace/pred'

I got the duplicate result in '/tmp/starspace/pred' as follows:

Example 0:
LHS:
, movie studios launch legal offensive against online pirates , los angeles - hollywood studios said thursday they will file hundreds of lawsuits later this month against individuals who swap pirated copies of movies over the internet .
RHS:
__label__4
Predictions:
(--) [0.164135] __label__4
(++) [0.164135] __label__4
(--) [0.0509765] __label__3
(--) [0.0509765] __label__3
(--) [-0.0315386] __label__1

Example 1:
LHS:
, super ant colony hits australia , a giant 100km colony of ants which has been discovered in melbourne , australia , could threaten local insect species .
RHS:
__label__4
Predictions:
(--) [0.0536099] __label__4
(++) [0.0536099] __label__4

I wonder if it's a bug or I just misunderstand something.

tusharkaley commented

Hey @ZhaoyangChen did you find a reason why this was happening or how to fix this?