Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

Unable to reproduce the result on STS tasks #124

Open
saharghannay opened this issue Apr 29, 2019 · 1 comment
Open

Unable to reproduce the result on STS tasks #124

saharghannay opened this issue Apr 29, 2019 · 1 comment

Comments

@saharghannay
Copy link

Hi,

I evaluate the InferSent 1 model on STS transfer tasks (STS12-16 and STSBenchmark) using the SentEval tool.
However I could not reproduce the same results.

2019-04-28 17:35:35,511 : ***** Transfer task : STS12 *****
Found 8104(/8122) words with w2v vectors
Vocab size : 8104
2019-04-28 17:35:37,836 : ALL (weighted average) : Pearson = 0.5893, Spearman = 0.5896
2019-04-28 17:35:37,836 : ALL (average) : Pearson = 0.5792, Spearman = 0.5798

2019-04-28 17:35:37,836 : ***** Transfer task : STS13 (-SMT) *****
Found 5106(/5116) words with w2v vectors
Vocab size : 5106
2019-04-28 17:35:38,933 : ALL (weighted average) : Pearson = 0.6399, Spearman = 0.6371
2019-04-28 17:35:38,933 : ALL (average) : Pearson = 0.5553, Spearman = 0.5541

2019-04-28 17:35:38,933 : ***** Transfer task : STS14 *****
Found 9046(/9060) words with w2v vectors
Vocab size : 9046
2019-04-28 17:35:41,391 : ALL (weighted average) : Pearson = 0.6545, Spearman = 0.6377
2019-04-28 17:35:41,391 : ALL (average) : Pearson = 0.6411, Spearman = 0.6258

2019-04-28 17:35:41,391 : ***** Transfer task : STS15 ****
Found 7313(/7335) words with w2v vectors
Vocab size : 7313
2019-04-28 17:35:43,532 : ALL (weighted average) : Pearson = 0.6879, Spearman = 0.6988
2019-04-28 17:35:43,533 : ALL (average) : Pearson = 0.6600, Spearman = 0.6731

2019-04-28 17:35:43,533 : ***** Transfer task : STS16 *****
Found 3930(/3971) words with w2v vectors
Vocab size : 3930
2019-04-28 17:35:44,572 : ALL (weighted average) : Pearson = 0.6287, Spearman = 0.6465
2019-04-28 17:35:44,572 : ALL (average) : Pearson = 0.6321, Spearman = 0.6494

***** Transfer task : STSBenchmark*****
Found 15718(/15755) words with w2v vectors
Vocab size : 15718
2019-04-28 17:36:19,249 : Test : Pearson 0.68976421922 Spearman 0.683016070529

PS: I used /dataset/get_data.bash to download the data,
for word embeddings I was using CBOW trained on wiki data.

Is there something wrong with the default setting?

Thanks for your help!

@saharghannay
Copy link
Author

Hi, I want to correct something :
I download the data with get_transfer_data.bash from SenEval.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant