valid_data = H5PYDataset(DATASET_PATH,
which_sets=('valid', ),
sources=('s_transition_obs', 'r_transition_obs',
'obs', 'actions'))
stream_valid = DataStream(valid_data,
iteration_scheme=SequentialScheme(
valid_data.num_examples, batch_size))
net = LstmSimpleNet2Pusher(27, 6)
print(net)
if CUDA:
net.cuda()
viz = VisdomExt([["loss", "validation loss"], ["diff"]], [
dict(title='LSTM loss', xlabel='iteration', ylabel='loss'),
dict(title='Diff loss', xlabel='iteration', ylabel='error')
])
means = {
'o':
np.array([
2.2281456, 1.93128324, 1.63007331, 0.48472479, 0.4500702, 0.30325469,
-0.38825685, -0.63075501, 0.63863981, -0.63173348, 1.01628101,
-1.02707994
],
dtype='float32'),
's':
np.array([
2.25090551, 1.94997263, 1.6495719, 0.43379614, 0.3314755, 0.43763939,
-0.38825685, -0.63075501, 0.63863981, -0.63173348, 1.01628101,
-1.02707994
net = torch.nn.Sequential(
torch.nn.Linear(15, HIDDEN_NODES),
nn.LeakyReLU(0.2),
torch.nn.Linear(HIDDEN_NODES, HIDDEN_NODES),
nn.LeakyReLU(0.2),
torch.nn.Linear(HIDDEN_NODES, HIDDEN_NODES/2),
nn.LeakyReLU(0.2),
torch.nn.Linear(HIDDEN_NODES/2, 6),
)
print(net)
if CUDA:
net.cuda()
viz = VisdomExt([["loss", "validation loss"],["diff"]],[dict(title='LSTM loss', xlabel='iteration', ylabel='loss'),
dict(title='Diff loss', xlabel='iteration', ylabel='error')])
def makeIntoVariables(dat):
input_ = np.concatenate([dat["obs"][:,:,:6], dat["actions"], dat["s_transition_obs"][:,:,:6]], axis=2)
x, y = autograd.Variable(
# Don't predict palet and goal position
torch.from_numpy(input_).cuda(),
requires_grad=False
), autograd.Variable(
torch.from_numpy(dat["r_transition_obs"][:,:,:6]).cuda(),
requires_grad=False
)
return x, y