Error increasing with new epochs #59
Description
Hi, I have been trying to do an sentiment analysis on twitter data using sswe. However, as each epoch is passing, the error value is continuously increasing and accuracy always remains 0. Can someone explain to me why this is happening!!
The following is a sample of error:
Epoch: 39, pairs: 900000, sent: 406, avg. error: 1794.635
Epoch: 39, pairs: 910000, sent: 919, avg. error: 1811.219
40 epochs Examples: 914760 Error: 124025225.271232 Accuracy: 0.000000 11924 corrections skipped
Epoch: 40, pairs: 920000, sent: 259, avg. error: 1841.875
Epoch: 40, pairs: 930000, sent: 772, avg. error: 1851.828
41 epochs Examples: 937629 Error: 126510924.248558 Accuracy: 0.000000 11738 corrections skipped
Epoch: 41, pairs: 940000, sent: 117, avg. error: 1883.111
Epoch: 41, pairs: 950000, sent: 619, avg. error: 1891.424
Epoch: 41, pairs: 960000, sent: 1141, avg. error: 1913.427
42 epochs Examples: 960498 Error: 129020667.139668 Accuracy: 0.000000 11861 corrections skipped
Epoch: 42, pairs: 970000, sent: 474, avg. error: 1932.033
I am using python dl-sentiwords.py training_twitter.tsv --vocab twitter_words.txt --vectors twitter_word_vectors.txt --variant word2vec
command to run sswe. The tsv file contains actual tweets in this format:
619971047195045888 1 negative "if these runway renovations at the airport prevent me from seeing taylor swift on monday, bad blood will have a new meaning."
words.txt contains the individual words and vectors.txt contains word2vec embedding for each word.
Any help will be highly appreciated.