more options #4
Description
implement
- max_df,
- think about lowercasing (done by default by vectorizer),
- json output for predictions?
- mlp num hidden layers, hidden layer sizes
- activation (although relu most likely is the best)
- solver (although: if not adam or sgd, then early stopping cannot be used)
- mlp alpha (l2, weight decay)