Fix go get

Fix go get tag

Add average across categories to SparseCategoricalAccuracy

Fix issue when reading metric logs
Merge fix for LcuDNNLTSM for newer versions of tensorflow

Add compile option to save model for cpu inference for CuDNNLSTM laye…

…r usage

Add example for training on gpu and saving for cpu inference

Add cuDNN accelerated LSTM layer

Speed up training by defining the shape of class weights
Update jobs example to use new CuDNNLSTM layer for massive speed increase

Improve performance with optional batchSize in compile call

Fix bug with yProcessor not getting saved
Add benchmark example
Change signature of CompileAndLoad to allow changes in future without signature change
Use correct default for embedding layer initializer
Improve performance of fit by moving metrics and callbacks into go func

Add MSE loss and ability for non-categorical model types

Add configurable y processor for single file and values dataset
Add helper methods for binary/sparse categorical raw/tokenized y processors
Add concurrency to values dataset to speed things up
Update examples with new y processor design
Fix example output of jobs
Remove testing monet example
Update transfer learning example with new weights get/set design
Add LeakyReLU layer
Add model methods to get weights in correct order / specific layer weights
Remove hardcoded [][]int32 yTrue and [][]float32 yPred casts to allow for other model types
Make categorical tokenizers have unlimited numWords
Add early stopping callback for when a metric flatlines

Add loading vanilla Tensorflow/Keras models

Add vanilla example
Add logic to validate we have a genuine TFKG model in model.LoadMoodel
Fix bug where inaccurate dataset len or an error were causing the callback.EventEnd not to fire on train/val/test

Add transfer learning functionality between TFKG models with .SetLaye…

…rWeights() on layer interface

Fix issue where the first row of single_file_dataset was being ignored
Add transfer_learning example
Save order of model weights in json file in save dir to make transfer learning nicer

Fix old version of python_generated.go being committed

Capture popular tokens better with tokenizer