Implement Pre-Built Customer Reviews Classifier Model [resolves #232]

[ertugrul-dmr]

sadedegel/prebuilt/README.md:

• Added how to use prebuilt pipeline
• Added how to use evaluation
• Added holdout test set f-1 macro score

sadedegel/prebuilt/model/customer_review_classification.joblib:

• Dumped pretrained/optimized model weights

sadedegel/prebuilt/customer_reviews_classification.py:

• Updated functions according to CONTRIBUTING.md
• Added, build, load and evaluate functions
• Added option to check roc_auc score for predict_proba method
• Model and vectorizer parameters are selected using optimizer with 3 fold results for 50 combinations

tests/prebuilt/context.py:

• Added import line for the base model

tests/prebuilt/test_movie_review_sentiment.py:

• Added assertions for testing purposes.

To-do:

• Going to update data paths for sadedegel dataset imports when they're available. (Self note, clean the reminder comments after that)

[husnusensoy]

Where is the dataset ?

[ertugrul-dmr]

I sent it to @dafajon, I will be implementing imports accordingly when it's up

[dafajon]

import as customer_review_classification. I implemented with this name.

[husnusensoy]

Do we have any baseline score to compare our model(s) that we can cite in README ?

[ertugrul-dmr]

The page I got the dataset from claims 84% F-1 score here: https://www.kaggle.com/savasy/multiclass-classification-data-for-turkish-tc32

[husnusensoy]

document it as we did in other models. Properly site it and resend you PR