I got it to work now, by getting the latest code from the git repo. But I am still a bit puzzled about why. The first code below fails in iteration 2 on loss = criterion(pred[:, :-future], test_target)
from __future__ import print_function
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
class Sequence(nn.Module):
def __init__(self):
super(Sequence, self).__init__()
self.lstm1 = nn.LSTMCell(1, 51)
self.lstm2 = nn.LSTMCell(51, 51)
self.linear = nn.Linear(51, 1)
def forward(self, input, future = 0):
outputs = []
h_t = Variable(torch.zeros(input.size(0), 51).double(), requires_grad=False)
c_t = Variable(torch.zeros(input.size(0), 51).double(), requires_grad=False)
h_t2 = Variable(torch.zeros(input.size(0), 51).double(), requires_grad=False)
c_t2 = Variable(torch.zeros(input.size(0), 51).double(), requires_grad=False)
for i, input_t in enumerate(input.chunk(input.size(1), dim=1)):
h_t, c_t = self.lstm1(input_t, (h_t, c_t))
h_t2, c_t2 = self.lstm2(h_t, (h_t2, c_t2))
output = self.linear(h_t2)
outputs += [output]
for i in range(future):# if we should predict the future
h_t, c_t = self.lstm1(output, (h_t, c_t))
h_t2, c_t2 = self.lstm2(h_t, (h_t2, c_t2))
output = self.linear(h_t2)
outputs += [output]
outputs = torch.stack(outputs, 1).squeeze(2)
return outputs
if __name__ == '__main__':
# set random seed to 0
np.random.seed(0)
torch.manual_seed(0)
# load data and make training set
data = torch.load('traindata.pt')
input = Variable(torch.from_numpy(data[3:, :-1]), requires_grad=False)
target = Variable(torch.from_numpy(data[3:, 1:]), requires_grad=False)
test_input = Variable(torch.from_numpy(data[:3, :-1]), requires_grad=False)
test_target = Variable(torch.from_numpy(data[:3, 1:]), requires_grad=False)
# build the model
seq = Sequence()
seq.double()
criterion = nn.SmoothL1Loss()
# use LBFGS as optimizer since we can load the whole data to train
optimizer = optim.LBFGS(seq.parameters(), lr=0.8)
#begin to train
for i in range(15):
print('STEP: ', i)
def closure():
optimizer.zero_grad()
out = seq(input)
loss = criterion(out, target)
print('loss:', loss.data.numpy())
loss.backward()
return loss
optimizer.step(closure)
# begin to predict
future = 1000
pred = seq(test_input, future = future)
print('test_target',test_target.data.numpy())
print('future:',future)
print('pred:',pred.data.numpy())
loss = criterion(pred[:, :-future], test_target)
print('test loss:', loss.data.numpy())
y = pred.data.numpy()
# draw the result
plt.figure(figsize=(30,10))
plt.title('Predict future values for time sequences\n(Dashlines are predicted values)', fontsize=30)
plt.xlabel('x', fontsize=20)
plt.ylabel('y', fontsize=20)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
def draw(yi, color):
plt.plot(np.arange(input.size(1)), yi[:input.size(1)], color, linewidth = 2.0)
plt.plot(np.arange(input.size(1), input.size(1) + future), yi[input.size(1):], color + ':', linewidth = 2.0)
draw(y[0], 'r')
draw(y[1], 'g')
draw(y[2], 'b')
plt.savefig('predict%d.pdf'%i)
plt.close()
However, the latest code from the github repo, runs without issues:
from __future__ import print_function
import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
class Sequence(nn.Module):
def __init__(self):
super(Sequence, self).__init__()
self.lstm1 = nn.LSTMCell(1, 51)
self.lstm2 = nn.LSTMCell(51, 51)
self.linear = nn.Linear(51, 1)
def forward(self, input, future = 0):
outputs = []
h_t = torch.zeros(input.size(0), 51, dtype=torch.double)
c_t = torch.zeros(input.size(0), 51, dtype=torch.double)
h_t2 = torch.zeros(input.size(0), 51, dtype=torch.double)
c_t2 = torch.zeros(input.size(0), 51, dtype=torch.double)
for i, input_t in enumerate(input.chunk(input.size(1), dim=1)):
h_t, c_t = self.lstm1(input_t, (h_t, c_t))
h_t2, c_t2 = self.lstm2(h_t, (h_t2, c_t2))
output = self.linear(h_t2)
outputs += [output]
for i in range(future):# if we should predict the future
h_t, c_t = self.lstm1(output, (h_t, c_t))
h_t2, c_t2 = self.lstm2(h_t, (h_t2, c_t2))
output = self.linear(h_t2)
outputs += [output]
outputs = torch.stack(outputs, 1).squeeze(2)
return outputs
if __name__ == '__main__':
# set random seed to 0
np.random.seed(0)
torch.manual_seed(0)
# load data and make training set
data = torch.load('traindata.pt')
input = torch.from_numpy(data[3:, :-1])
target = torch.from_numpy(data[3:, 1:])
test_input = torch.from_numpy(data[:3, :-1])
test_target = torch.from_numpy(data[:3, 1:])
# build the model
seq = Sequence()
seq.double()
criterion = nn.MSELoss()
# use LBFGS as optimizer since we can load the whole data to train
optimizer = optim.LBFGS(seq.parameters(), lr=0.8)
#begin to train
for i in range(15):
print('STEP: ', i)
def closure():
optimizer.zero_grad()
out = seq(input)
loss = criterion(out, target)
print('loss:', loss.item())
loss.backward()
return loss
optimizer.step(closure)
# begin to predict, no need to track gradient here
with torch.no_grad():
future = 1000
pred = seq(test_input, future=future)
loss = criterion(pred[:, :-future], test_target)
print('test loss:', loss.item())
y = pred.detach().numpy()
# draw the result
plt.figure(figsize=(30,10))
plt.title('Predict future values for time sequences\n(Dashlines are predicted values)', fontsize=30)
plt.xlabel('x', fontsize=20)
plt.ylabel('y', fontsize=20)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
def draw(yi, color):
plt.plot(np.arange(input.size(1)), yi[:input.size(1)], color, linewidth = 2.0)
plt.plot(np.arange(input.size(1), input.size(1) + future), yi[input.size(1):], color + ':', linewidth = 2.0)
draw(y[0], 'r')
draw(y[1], 'g')
draw(y[2], 'b')
plt.savefig('predict%d.pdf'%i)
plt.close()
There are some slight differences, but I am not proficient enough in pyTorch to understand why they should make a difference.