Release v0.0.20

Make the number of hidden units adjustable

Release v0.0.19

reduce GPU memory && max/min normalization

• use numpy and numba instead of tensorflow in data
• use main memory in data module instead of GPU
• add normalization for large features

Release v0.0.18

Release v0.0.18

Make BN optional in AlphabetV4

Release v0.0.17

add another model variant AlphanetV4

Release v0.0.16

Release v0.0.16

allow choosing between GRU and LSTM

Release v0.0.15

Support TensorFlow 2.6.0

Release v0.0.14

allowing overriding val_length in data.TrainValData.get()

Release v0.0.13

data module full support for classification task

Release V0.0.12

Reduce memory usage in data tool

Release v0.0.11

improve data pipeline memory efficiency