Improve sentiment model accuracy and log loss

[dillonmabry]

Improve both basic and advanced ML models. Current accuracy is ~78% for basic model using Logistic Regression and hyperparams.

Test and select from the following:

• Twitter traditional ML model
• IMDB Conv1D tensorflow model
• Twitter trained Conv1D tensorflow model

[dillonmabry]

Current accuracy is ~80% for hypertuned model, validation holdout set similar accuracy.

Still working on possible tuning of a deeper model.

[dillonmabry]

Explored deep models and will stick with a more traditional model for generalized sentiment analysis. A deep model would be too specific to certain topics/applications (ie, Twitter, Youtube, etc.)