def post_epoch_visualize(self, epoch, split):
# if self.flags.visualize_only:
# self.do_plots()
# else:
if True:
print('* Visualizing', split)
Z = torch.linspace(0.0 + 1e-3, 1.0 - 1e-3, steps=20)
Z = torch.cartesian_prod(Z, Z).view(20, 20, 2)
if self.flags.z_size == 2 and self.flags.normal_latent:
dist = normal.Normal(0.0, 1.0)
x_gens = []
for row in range(20):
if self.flags.z_size == 2:
z = Z[row]
if self.flags.normal_latent:
z = dist.icdf(z)
else:
if self.flags.normal_latent:
z = torch.randn(20, self.flags.z_size)
else:
z = torch.rand(20, self.flags.z_size)
z = self.model.prepare_batch(z)
x_gen = self.model.run_batch([z]).view(20, self.img_chan, self.img_size, self.img_size).detach().cpu()
x_gens.append(x_gen)
x_full = torch.cat(x_gens, dim=0).numpy()
if split == 'test':
fname = self.flags.log_dir + '/test.png'
else:
fname = self.flags.log_dir + '/vis_%03d.png' % self.model.get_train_steps()
misc.save_comparison_grid(fname, x_full, border_width=0, retain_sequence=True)
print('* Visualizations saved to', fname)
def post_epoch_visualize(self, epoch, split):
print('* Visualizing', split)
fname = self.flags.log_dir + '/vis_{}_{}.png'.format(
self.model.get_train_steps(), split)
if split == 'train':
z = torch.randn(self.flags.batch_size, self.flags.latent_size, 1,
1)
z = self.model.prepare_batch(z)
x = self.model.run_batch([None, z], visualize=True)
vis_data = x.cpu().numpy()
aspect = 1.0
else:
batch = next(
self.reader.iter_batches(split,
self.batch_size,
shuffle=True,
partial_batching=True,
threads=self.threads,
max_batches=1))
batch = self.model.prepare_batch(batch[0])
x = self.model.run_batch([batch, None], visualize=True)
y = batch.cpu().numpy()[None, ...]
x = x.cpu().numpy()[None, ...]
vis_data = np.concatenate([y, x], axis=0)
vis_data = np.swapaxes(vis_data, 0, 1).reshape(-1, *x.shape[2:])
aspect = 2.0
misc.save_comparison_grid(fname,
vis_data,
border_shade=0.5,
retain_sequence=True,
desired_aspect=aspect)
print('* Visualizations saved to', fname)
def post_epoch_visualize(self, epoch, split):
if split != 'train':
print('* Visualizing', split)
vis_data, rows_cols = self._visualize_split(
split, min(self.flags.seq_len - 1, 10), 5)
if split == 'test':
fname = self.flags.log_dir + '/test.png'
else:
fname = self.flags.log_dir + '/val%03d.png' % epoch
misc.save_comparison_grid(fname,
vis_data,
border_shade=1.0,
rows_cols=rows_cols,
retain_sequence=True)
print('* Visualizations saved to', fname)
if split == 'test':
print('* Generating more visualizations for', split)
vis_data, rows_cols = self._visualize_split(split, 0, 15)
fname = self.flags.log_dir + '/test_more.png'
misc.save_comparison_grid(fname,
vis_data,
border_shade=1.0,
rows_cols=rows_cols,
retain_sequence=True)
print('* More visualizations saved to', fname)
def post_epoch_visualize(self, epoch, split):
print('* Visualizing', split)
out = self._visualize_split(split, self.history_length, 5)
if out is not None:
vis_data, aspect = out
fname = self.flags.log_dir + '/{}'.format(split) + '%03d.png' % epoch
misc.save_comparison_grid(fname, vis_data, rows_cols=aspect, border_shade=1.0, retain_sequence=True)
print('* Visualizations saved to', fname)
else:
print('* Visualization skipped')
def post_epoch_visualize(self, epoch, split):
print('* Visualizing', split)
length = min(10, self.flags.seq_len - 1)
n = min(5, self.flags.seq_len - length)
vis_data, aspect = self._visualize_split(split, length, 5)
fname = self.flags.log_dir + '/{}'.format(split) + '%03d.png' % epoch
misc.save_comparison_grid(fname,
vis_data,
rows_cols=aspect,
border_shade=1.0,
retain_sequence=True)
print('* Visualizations saved to', fname)
def post_epoch_visualize(self, epoch, split):
print('* Visualizing', split)
Z = torch.linspace(0.0, 1.0, steps=20)
Z = torch.cartesian_prod(Z, Z).view(20, 20, 2)
x_gens = []
for row in range(20):
if self.flags.z_size == 2:
z = Z[row]
else:
z = torch.rand(20, self.flags.z_size)
z = self.model.prepare_batch(z)
x_gen = self.model.run_batch([z], visualize=True).detach().cpu()
x_gens.append(x_gen)
x_full = torch.cat(x_gens, dim=0).numpy()
if split == 'test':
fname = self.flags.log_dir + '/test.png'
else:
fname = self.flags.log_dir + '/vis_%03d.png' % self.model.get_train_steps()
misc.save_comparison_grid(fname, x_full, border_width=0, retain_sequence=True)
print('* Visualizations saved to', fname)
from pylego.misc import save_comparison_grid
from readers.moving_mnist import MovingMNISTReader
if __name__ == '__main__':
reader = MovingMNISTReader('data/MNIST')
for i, batch in enumerate(reader.iter_batches('train', 4, max_batches=5)):
print(batch.size())
if i < 3:
batch = batch[0].numpy().reshape(20, 1, 28, 28)
save_comparison_grid('seq%d.png' % i, batch)
if __name__ == '__main__':
emulator = GymReader('Seaquest-v0', 6, 4, 2, 100)
reader = ReplayBuffer(emulator, 5000, 100, 1, 4, 0.99)
print('EMULATOR:')
for i, batch in enumerate(
emulator.iter_batches('train', 4, max_batches=5, threads=2)):
obs, actions, rewards, done = batch.get_next()[:4]
print('obs', obs.size())
print('actions', actions.shape)
print('rewards', rewards.shape)
print()
if i < 3:
batch = obs.numpy().reshape(obs.shape[0] * obs.shape[1], 3, 80, 80)
save_comparison_grid('eseq%d.png' % i,
batch,
rows_cols=obs.shape[:2],
retain_sequence=True)
print(actions)
print()
print('REPLAY BUFFER READER:')
for i, batch in enumerate(reader.iter_batches('train', 4, max_batches=5)):
obs, actions, rewards, done, t1, t2, returns, is_weight, idx = batch
print('obs', obs.size())
print('actions', actions.shape)
print('rewards', rewards.shape)
print('is_weight', is_weight.shape)
print()
if i < 3:
batch = obs.numpy().reshape(4 * 6, 3, 80, 80)
save_comparison_grid('rseq%d.png' % i,
from pylego.misc import save_comparison_grid
from readers.cifar10 import CIFAR10Reader
if __name__ == '__main__':
reader = CIFAR10Reader('data')
for i, batch in enumerate(reader.iter_batches('train', 256,
max_batches=5)):
img = batch[0].numpy()
print(img.shape, img.min(), img.max())
save_comparison_grid('seq%d.png' % i, img, border_shade=0.75)
with open(batches_fname, 'wb') as f:
pickle.dump(batches, f)
else:
with open(batches_fname, 'rb') as f:
batches = pickle.load(f)
return batches
if __name__ == '__main__':
try:
os.makedirs(DATA_DIR)
except IOError as e:
pass
batches = get_batches(DATA_DIR + '/norm_batches.pk')
misc.save_comparison_grid('example1.png', batches[:16], border_shade=0.8)
h, w = batches.shape[2:]
crop_top, crop_bottom, crop_left, crop_right = 34, 16, 0, 0
print(batches.shape)
batches = batches[:, :, crop_top:h-crop_bottom, crop_left:w-crop_right]
print(batches.shape)
flat_batches = batches.transpose(1, 0, 2, 3).reshape(batches.shape[1], -1)
mean = flat_batches.mean(axis=1)[None, :, None, None]
std = flat_batches.std(axis=1)[None, :, None, None]
batches -= mean
batches /= std
batches = batches.mean(axis=1, keepdims=True) # greyscale
true_min = batches.min()
true_max = batches.max()
print(true_min, true_max)
bmin = true_min # np.percentile(batches, 0)