lstm04.py

input_size  = 50  # representing the one-hot encoded vector size
hidden_size = 100 # number of hidden nodes in the LSTM layer
n_layers    = 2   # number of LSTM layers
output_size = 50  # output of 50 scores for the next character

lstm   = nn.LSTM(input_size, hidden_size, n_layers, batch_first=True)
linear = nn.Linear(hidden_size, output_size)

# Data Flow Protocol
# 1. network input shape: (batch_size, seq_length, num_features)
# 2. LSTM output shape: (batch_size, seq_length, hidden_size)
# 3. Linear input shape:  (batch_size * seq_length, hidden_size)
# 4. Linear output: (batch_size * seq_length, out_size)

x = get_batches(data)         
x, hs = lstm(x, hs)
x = x.reshape(-1, hidden_size) 
x = linear(x)