1-RNN-basics.ipynb

Kernel: Python 3

import torch
import numpy as np

# Random seed to make results deterministic and reproducible
torch.manual_seed(0)

<torch._C.Generator at 0x110204b50>

# declare dimension
input_size = 4
hidden_size = 2

# singleton example
# shape : (1, 1, 4)
# input_data_np = np.array([[[1, 0, 0, 0]]])

# sequential example
# shape : (3, 5, 4)
h = [1, 0, 0, 0]
e = [0, 1, 0, 0]
l = [0, 0, 1, 0]
o = [0, 0, 0, 1]
input_data_np = np.array([[h, e, l, l, o], [e, o, l, l, l], [l, l, e, e, l]], dtype=np.float32)

# transform as torch tensor
input_data = torch.Tensor(input_data_np)

# declare RNN
rnn = torch.nn.RNN(input_size, hidden_size)

# check output
outputs, _status = rnn(input_data)
print(outputs)
print(outputs.size())

tensor([[[-0.7497, -0.6135],
         [-0.5282, -0.2473],
         [-0.9136, -0.4269],
         [-0.9136, -0.4269],
         [-0.9028,  0.1180]],

        [[-0.5753, -0.0070],
         [-0.9052,  0.2597],
         [-0.9173, -0.1989],
         [-0.9173, -0.1989],
         [-0.8996, -0.2725]],

        [[-0.9077, -0.3205],
         [-0.8944, -0.2902],
         [-0.5134, -0.0288],
         [-0.5134, -0.0288],
         [-0.9127, -0.2222]]], grad_fn=<StackBackward>)
torch.Size([3, 5, 2])