2martens/masterthesis
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Removed visualization of latent space

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-04-17 14:14:27 +02:00
parent bc07fda615
commit 600cb55e2e
2 changed files with 16 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
src/twomartens/masterthesis/aae/run.py
View File

@ -97,15 +97,14 @@ def _run_one_epoch_simple(dataset: tf.data.Dataset,
dtype=tf.float32) dtype=tf.float32)
for x in dataset: for x in dataset:
reconstruction_loss, x_decoded, z = _run_enc_dec_step_simple(encoder=encoder, reconstruction_loss, x_decoded = _run_enc_dec_step_simple(encoder=encoder,
decoder=decoder, decoder=decoder,
inputs=x, inputs=x,
global_step=global_step) global_step=global_step)
enc_dec_loss_avg(reconstruction_loss) enc_dec_loss_avg(reconstruction_loss)
if int(global_step % train.LOG_FREQUENCY) == 0: if int(global_step % train.LOG_FREQUENCY) == 0:
comparison = K.concatenate([x[:int(batch_size / 2)], x_decoded[:int(batch_size / 2)], comparison = K.concatenate([x[:int(batch_size / 2)], x_decoded[:int(batch_size / 2)]], axis=0)
z[:int(batch_size / 2)]], axis=0)
grid = util.prepare_image(comparison.cpu(), nrow=int(batch_size / 2)) grid = util.prepare_image(comparison.cpu(), nrow=int(batch_size / 2))
summary_ops_v2.image(name='reconstruction', summary_ops_v2.image(name='reconstruction',
tensor=K.expand_dims(grid, axis=0), max_images=1, tensor=K.expand_dims(grid, axis=0), max_images=1,
@ -126,7 +125,7 @@ def _run_one_epoch_simple(dataset: tf.data.Dataset,
def _run_enc_dec_step_simple(encoder: model.Encoder, decoder: model.Decoder, def _run_enc_dec_step_simple(encoder: model.Encoder, decoder: model.Decoder,
inputs: tf.Tensor, inputs: tf.Tensor,
global_step: tf.Variable) -> Tuple[tf.Tensor, tf.Tensor, tf.Tensor]: global_step: tf.Variable) -> Tuple[tf.Tensor, tf.Tensor]:
""" """
Runs the encoder and decoder jointly for one step (one batch). Runs the encoder and decoder jointly for one step (one batch).
@ -148,8 +147,4 @@ def _run_enc_dec_step_simple(encoder: model.Encoder, decoder: model.Decoder,
summary_ops_v2.scalar(name='reconstruction_loss', tensor=reconstruction_loss, summary_ops_v2.scalar(name='reconstruction_loss', tensor=reconstruction_loss,
step=global_step) step=global_step)
input_shape = tf.shape(inputs) return reconstruction_loss, x_decoded
z_reshaped = tf.reshape(z, [-1, input_shape[1], input_shape[2], 1])
z_concatenated = K.concatenate((z_reshaped, z_reshaped, z_reshaped), axis=3)
return reconstruction_loss, x_decoded, z_concatenated

23
src/twomartens/masterthesis/aae/train.py
View File

@ -161,17 +161,16 @@ def _train_one_epoch_simple(epoch: int,
print("learning rate change!") print("learning rate change!")
for x in dataset: for x in dataset:
reconstruction_loss, x_decoded, z = _train_enc_dec_step_simple(encoder=encoder, reconstruction_loss, x_decoded = _train_enc_dec_step_simple(encoder=encoder,
decoder=decoder, decoder=decoder,
optimizer=enc_dec_optimizer, optimizer=enc_dec_optimizer,
inputs=x, inputs=x,
global_step_enc_dec=global_step_enc_dec, global_step_enc_dec=global_step_enc_dec,
global_step=global_step) global_step=global_step)
enc_dec_loss_avg(reconstruction_loss) enc_dec_loss_avg(reconstruction_loss)
if int(global_step % LOG_FREQUENCY) == 0: if int(global_step % LOG_FREQUENCY) == 0:
comparison = K.concatenate([x[:int(batch_size / 2)], x_decoded[:int(batch_size / 2)], comparison = K.concatenate([x[:int(batch_size / 2)], x_decoded[:int(batch_size / 2)]], axis=0)
z[:int(batch_size/2)]], axis=0)
grid = util.prepare_image(comparison.cpu(), nrow=int(batch_size/2)) grid = util.prepare_image(comparison.cpu(), nrow=int(batch_size/2))
summary_ops_v2.image(name='reconstruction', summary_ops_v2.image(name='reconstruction',
tensor=K.expand_dims(grid, axis=0), max_images=1, tensor=K.expand_dims(grid, axis=0), max_images=1,
@ -194,8 +193,7 @@ def _train_enc_dec_step_simple(encoder: model.Encoder, decoder: model.Decoder,
optimizer: tf.train.Optimizer, optimizer: tf.train.Optimizer,
inputs: tf.Tensor, inputs: tf.Tensor,
global_step: tf.Variable, global_step: tf.Variable,
global_step_enc_dec: tf.Variable, global_step_enc_dec: tf.Variable) -> Tuple[tf.Tensor, tf.Tensor]:
debug: bool) -> Tuple[tf.Tensor, tf.Tensor, tf.Tensor]:
""" """
Trains the encoder and decoder jointly for one step (one batch). Trains the encoder and decoder jointly for one step (one batch).
@ -229,11 +227,8 @@ def _train_enc_dec_step_simple(encoder: model.Encoder, decoder: model.Decoder,
optimizer.apply_gradients(zip(enc_dec_grads, optimizer.apply_gradients(zip(enc_dec_grads,
encoder.trainable_variables + decoder.trainable_variables), encoder.trainable_variables + decoder.trainable_variables),
global_step=global_step_enc_dec) global_step=global_step_enc_dec)
input_shape = tf.shape(inputs)
z_reshaped = tf.reshape(z, [-1, input_shape[1], input_shape[2], 1])
z_expanded = K.concatenate((z_reshaped, z_reshaped, z_reshaped), axis=3)
return reconstruction_loss, x_decoded, z_expanded return reconstruction_loss, x_decoded
if __name__ == "__main__": if __name__ == "__main__":