Attached are the include files for the LSTM. The files included are:

• Gates - for the 4 gates used in an LSTMs. 
• TimeStep - which combines the gates, and in practical usage would represent the time series bars.
• LSTMNetwork - implementing the learning algorithms.

Also included is an example LSTMTest script using the Simple RPC indicator, also attached.

To create a new LSTM network, provide the constructor with number of patterns, number of inputs (predictors per timestep) and the number of timesteps, as shown below;

CLSTMNetwork *net=new LSTMNetwork(patterns,inputs,timesteps);

To teach the network, call the Learn function, providing it with the input array, the targets array, the learning error threshold, and the number of learning epochs as below;

net.Learn(in,tg,mse,500000);

After learning, the final error and epochs taken to converge can be acquired from the network as below;

net.MSE();  net.Epochs();

To check a particular pattern against the network, the Calculate function is called, passing the candidate pattern into the function as a parameter as shown;

net.Calculate(in);

The Calculate function returns the output. This LSTM has a single output neuron.

If anyone finds bugs or has improvements or any suggestions, please be kind enough to share. Good luck.