I have completed the code that predict stock market with LSTM neural network. I have a problem in giving and receiving data from a neural network. I have minibatch, each minibatch have two arrays: the input price, and output ideal. Under the input price I mean that it has 'n' ticks, under output ideal I mean stochastic indicator, which has a long period, with a shift=-100. Ie, NN turns out in "probabilistic" neural network.
Well, I have a history with the 1000 tick data. Configuration: minibatch = 1, input = 1 (above I called it "'n' ticks"), output = 1.
As soon as I train a neural network, from [0; 1000) ticks are served on the neural network only [0, 900), because the indicator data over. How to get a prediction for last tick? Just ignore the interval [900, 1000) and immediately apply the last (1000th) tick??
Arrays for minibatch calculates by this form:
for(i=0, i<minibatch_size, i=i+1)
input[i] = if (i>0) tick[i]-tick[i-1]; else 0;
ideal[0] = indicator[batch_size-1];
NN used tanh of the input and output layer.
I need to increase input up to 100 or output up to 100?? How to avoid this?
