Retrain and update IMDB sentiment analysis model

[jessijzhao]

• changed number of embedding dimensions from 50 to 100
• retrained model in models/imdb-model-cnn.pt
• ran IMDB_TorchText_Interpret.ipynb with new model and current library versions
• updated img/sentiment_analysis.png with image of results
• addresses issues Unproduce the results of text classification tutorial #399 and Questions about the IMDB sentiment analysis tutorial #547

[NarineK]

Thank you for working on this PR, @jessijzhao!
A question: It looks like the size of the model decreased by ~2x. Is that because we store 2x less parameters or did the old version store more than the parameters ?

[jessijzhao]

Thanks for the comment @NarineK! The model I trained has 2,620,801 parameters (matching the original training script by bentrevett) and the old model has 5,159,701. I followed the original script by keeping
MAX_VOCAB_SIZE = 25_000 whereas it looks like the old model was using all 101982 words - should I retrain the model with the full vocabulary?

[NarineK]

Thanks for the comment @NarineK! The model I trained has 2,620,801 parameters (matching the original training script by bentrevett) and the old model has 5,159,701. I followed the original script by keeping
MAX_VOCAB_SIZE = 25_000 whereas it looks like the old model was using all 101982 words - should I retrain the model with the full vocabulary?

Thank you for verifying it, @jessijzhao! If it is not too time consuming, let's try full vocab size. We might obtain better performance results.

[facebook-github-bot]

@NarineK has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@NarineK merged this pull request in 3c47914.