Add RNN, LSTM

[josephjaspers]

No description provided.

[xinsuinizhuan]

yes。lstm and rnn is best method to forecast! You are best! I am your fans!

[xinsuinizhuan]

Add RNN, LSTM?? whether to support CNN and LSTM？？

[josephjaspers]

Completed RNN, Working on LSTM next.

306872c
0916fa4
2377260

Example here:
https://github.com/josephjaspers/BlackCat_Tensors/tree/master/examples/mnist_test_recurrent

TODO (Not completed)
Add example using a better dataset.
Add LSTM, Add GRU
Add Convolution/CNN

[josephjaspers]

Added LSTM (experimental)
58ae986

LSTM is experimental.
I need to check and make sure the math for it is correct, it seems to be relatively working with the given example, though it performs worse than the plain recurrent layer.

Update:
Fixed a few bugs as of:
777818e

Seems to be working as expected now.

[xinsuinizhuan]

OK, thats in part good news。Others, i hope the trained modle could save, and when need, i only load model to predict.

[josephjaspers]

I will definitely add that soon, if you could give me a list of features you want I will try to implement them as fast as possible.

[xinsuinizhuan]

I will definitely add that soon, if you could give me a list of features you want I will try to implement them as fast as possible.

I want use the net to forecast the short-term-load for Electric-Power,as the net at https://github.com/YuanSiping/Short-Term-Load-Forecasting-for-Electric-Power-Systems
main supprot net:
1、cnn
2、lstm

But, i need the trained net support save and load net,as the net at https://github.com/Artelnics/OpenNN.
main supprot function:
1、save the trained model
2、load the saved trained model to predit
3、the best support the net could continue trained.

[josephjaspers]

Closing this issue as RNN/LSTM have been added.

Added a ticket for training/saving here:
#25