Implement Pre-Built Movie Review Sentiment Classifier [resolves #218]

[ertugrul-dmr]

sadedegel/prebuilt/README.md:

• Added how to use prebuilt pipeline
• Added cv score
• Added test set score

sadedegel/prebuilt/model/movie_sentiment.joblib:

• Dumped pretrained model weights

sadedegel/prebuilt/movie_reviews.py:

• Added cv, build, load and evaluate functions
• Model and vectorizer parameters selected using optimizer with 3 fold results.

tests/prebuilt/context.py:

• Added import line for the base model

tests/prebuilt/test_movie_review_sentiment.py:

• Added assertions for testing purposes.

[ertugrul-dmr]

I sent the training and test datasets to @dafajon today. Code should be ready to pull when data is added.

[husnusensoy]

• [ ] Use Text2Doc transformation in pipeline
• [ ] can you reimplement the model using icu tokenizer.
• [ ] Use a predictor with predict_proba
• [ ] Don't use cv in reporting your accuracy if you already have a seperate dataset
• [ ] Don't use partial_fit if your dataset is small enough.

[ertugrul-dmr]

Thanks for updating me, I'm already on it.

One question though, what would be the limit for small dataset? You have any number in your mind?

[husnusensoy]

Of course no I dont :) But for majority of the problems ~30K training instance can be used for training without mini batch. Please do check

• Profanity (pipeline)
• Sentiment (pipeline)
• News Categorizer (onlinepipeline)

[husnusensoy]

@ertugrul-dmr waiting for your prebuilt model confirming sadedegel/prebuilt/CONTRIBUTING.md conventions