sequence_train.py

import torch
from torch import optim
import torch.nn.functional as F
import generate_data
from utils import to_var
from pointer_network import PointerNetwork

total_size = 10000
weight_size = 256
emb_size = 32
batch_size = 250
n_epochs = 5

input_seq_len = 4
input, targets = generate_data.make_seq_data(total_size, input_seq_len)
inp_size = input_seq_len

# Convert to torch tensors
input = to_var(torch.LongTensor(input))     # (N, L)
targets = to_var(torch.LongTensor(targets)) # (N, L)

data_split = (int)(total_size * 0.9)
train_X = input[:data_split]
train_Y = targets[:data_split]
test_X = input[data_split:]
test_Y = targets[data_split:]


# from pointer_network import PointerNetwork
def train(model, X, Y, batch_size, n_epochs):
    model.train()
    optimizer = optim.Adam(model.parameters())
    N = X.size(0)
    L = X.size(1)
    # M = Y.size(1)
    for epoch in range(n_epochs + 1):
        # for i in range(len(train_batches))
        for i in range(0, N-batch_size, batch_size):
            x = X[i:i+batch_size] # (bs, L)
            y = Y[i:i+batch_size] # (bs, M)

            probs = model(x) # (bs, M, L)
            outputs = probs.view(-1, L) # (bs*M, L)
            # outputs = probs.view(L, -1).t().contiguous() # (bs*M, L)
            y = y.view(-1) # (bs*M)
            loss = F.nll_loss(outputs, y)

            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

        if epoch % 2 == 0:
            print('epoch: {}, Loss: {:.5f}'.format(epoch, loss.item()))
            # for _ in range(2): # random showing results
            #     pick = np.random.randint(0, batch_size)
            #     probs = probs.contiguous().view(batch_size, M, L).transpose(2, 1) # (bs, L, M)
            #     y = y.view(batch_size, M)
            #     print("predict: ", probs.max(1)[1].data[pick][0], probs.max(1)[1].data[pick][1],
            #           "target  : ", y.data[pick][0], y.data[pick][1])
            test(model, X, Y)


def test(model, X, Y):
    probs = model(X) # (bs, M, L)
    _v, indices = torch.max(probs, 2) # (bs, M)
    # show test examples
    # for i in range(len(indices)):
    #     print('-----')
    #     print('test', [v for v in X[i].data])
    #     print('label', [v for v in Y[i].data])
    #     print('pred', [v for v in indices[i].data])
    #     if torch.equal(Y[i].data, indices[i].data):
    #         print('eq')
    #     if i>20: break
    correct_count = sum([1 if torch.equal(ind.data, y.data) else 0 for ind, y in zip(indices, Y)])
    print('Acc: {:.2f}% ({}/{})'.format(correct_count/len(X)*100, correct_count, len(X)))

model = PointerNetwork(inp_size, emb_size, weight_size, input_seq_len)
if torch.cuda.is_available():
    model.cuda()
train(model, train_X, train_Y, batch_size, n_epochs)
print('----Test result---')
test(model, test_X, test_Y)