Multiple Input Streams

[GoatedChopin]

Pokemon Health Regression Model CNN + Stats:
https://www.youtube.com/watch?v=4-O14gOdRso

This video is a great tutorial on how to build a keras model with mixed_data, which could easily be adapted to this use-case. Although his pokemon health-regressor model predicts a continuous variable, changing the output layer should be sufficient alteration.

[GoatedChopin]

Implement checkpoints that save the best model using tensorflow.keras.callbacks.ModelCheckpoint

found at 45:10 in Greg Hogg's video. That should save on computation time if a certain number of epochs is best.

[GoatedChopin]

change final activation layer to softmax if not already
change loss function to categorical_crossentropy if not already.
https://datascience.stackexchange.com/questions/105101/which-keras-metric-for-multiclass-classification