LSTM after 2nd encoder.

First working version of the BERT classifier for sentence quality.

Fit model.

Load data.

Load BERT model.

Sentence quality classification with BERT.

Update sentence quality classification with new training data.

Cleanup.

GRU, batch size = 32, identity init in embedding layer.

Clean up.

LSTM character level model.

Input preparation for LSTM: One hot encoding character sequences.

Add early stopping.

Keras multilayer perceptron.

Long tail normalization.