seq2seq model in keras extended from addition to all four basic computations (+, -, *, /)
code taken from here: https://github.com/lukas/ml-class/blob/master/videos/seq2seq/train.py
original source seems to be: https://keras.io/examples/nlp/addition_rnn/
-
u need to define log- and checkpoint- path as well as log_file name in the 'NN_config.py'.
-
since I dont like using the keras FLAG stuff, this is my way to get global variables like path etc. using the global character of the imported variables.
-
at the end of each epoch u will get validation examples. to see how good it got so far. also the model will be saved at this point(3x). as it will be at the end of training.
-
during training there is a custom logfile written. it needs to be opened manually from another console. then its possible to watch the learning curves. (the matplotlib code is messy. sry, its old. I just quickly adapted it for this project.)
-
new training batches are being created before every epoch now, ..my conclusion is, that way, it cannot overfit so easily.
-
the current model takes quite some time to train. I am still experimenting with 'val_accuracy=0.9200' atm. so not every answer is correct. in my eyes it needs val_accuracy 0.99 at least to be trusted.
whats missing is user input to manually type in questions and hit enter for answer(s). feel free to test. have fun.