public static void Demo()
{
LSTMNetwork network = LSTMNetwork.Create2(32, 5);
network.DisplayByDot2();
Dictionary <string, double[]> inputs = new Dictionary <string, double[]>();
inputs.Add("inputs", new double[] { 1, 2, 3, 4, 5 });
network.Run2(inputs);
}
private static void Main(string[] args)
{
Console.Clear();
ProcessingDevice.Device = DeviceType.CPU;
OpenText();
hidden_size = 100;
seq_length = 25;
learning_rate = 1e-1f;
net = new LSTMNetwork(vocab_size, vocab_size, hidden_size, learning_rate, 1e-1f);
var hprev = new FloatArray(hidden_size);
var cprev = new FloatArray(hidden_size);
var smooth_loss = -Math.Log(1.0 / vocab_size) * seq_length;
int n = 0;
int p = 0;
while (n <= 1000 * 100)
{
if (p + seq_length + 1 >= data_size || n == 0)
{
hprev = new FloatArray(hidden_size);
cprev = new FloatArray(hidden_size);
p = 0;
}
var inputs = new int[seq_length];
var targets = new int[seq_length];
for (int i = 0; i < seq_length; i++)
{
inputs[i] = char_to_ix[txt[p + i]];
}
for (int i = 0; i < seq_length; i++)
{
targets[i] = char_to_ix[txt[p + 1 + i]];
}
(var loss, var dWf, var dWi, var dWc, var dWo, var dWv,
var dBf, var dBi, var dBc, var dBo, var dBv,
var hs, var cs) = net.BPTT(inputs, targets, hprev, cprev);
net.UpdateParams(dWf, dWi, dWc, dWo, dWv, dBf, dBi, dBc, dBo, dBv);
if (n % 100 == 0)
{
Sample(hprev, cprev, inputs[0], 200);
Console.WriteLine($"iter {n}, loss: {smooth_loss}");
}
hprev = hs;
cprev = cs;
smooth_loss = smooth_loss * 0.999 + loss * 0.001;
p += seq_length;
n += 1;
}
}
