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

Improved naming for histogram summaries and added variable summaries

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-02-08 20:47:34 +01:00
parent 0cbab44f01
commit a51b535728
1 changed files with 16 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/twomartens/masterthesis/aae/train.py
View File

@ -449,8 +449,10 @@ def _train_xdiscriminator_step(x_discriminator: XDiscriminator, decoder: Decoder
step=global_step) step=global_step)
summary_ops_v2.scalar(name='x_discriminator_loss', tensor=_xd_train_loss, summary_ops_v2.scalar(name='x_discriminator_loss', tensor=_xd_train_loss,
step=global_step) step=global_step)
for variable, grad in xd_grads: for grad, variable in zip(xd_grads, x_discriminator.trainable_variables):
summary_ops_v2.histogram(name='x_discriminator_grads/' + variable.name, tensor=tf.math.l2_normalize(grad), summary_ops_v2.histogram(name='gradients/' + variable.name, tensor=tf.math.l2_normalize(grad),
step=global_step)
summary_ops_v2.histogram(name='variables/' + variable.name, tensor=tf.math.l2_normalize(variable),
step=global_step) step=global_step)
optimizer.apply_gradients(zip(xd_grads, x_discriminator.trainable_variables), optimizer.apply_gradients(zip(xd_grads, x_discriminator.trainable_variables),
global_step=global_step_xd) global_step=global_step_xd)
@ -486,8 +488,10 @@ def _train_decoder_step(decoder: Decoder, x_discriminator: XDiscriminator,
if int(global_step % LOG_FREQUENCY) == 0: if int(global_step % LOG_FREQUENCY) == 0:
summary_ops_v2.scalar(name='decoder_loss', tensor=_decoder_train_loss, summary_ops_v2.scalar(name='decoder_loss', tensor=_decoder_train_loss,
step=global_step) step=global_step)
for variable, grad in grads: for grad, variable in zip(grads, decoder.trainable_variables):
summary_ops_v2.histogram(name='decoder_grads/' + variable.name, tensor=tf.math.l2_normalize(grad), summary_ops_v2.histogram(name='gradients/' + variable.name, tensor=tf.math.l2_normalize(grad),
step=global_step)
summary_ops_v2.histogram(name='variables/' + variable.name, tensor=tf.math.l2_normalize(variable),
step=global_step) step=global_step)
optimizer.apply_gradients(zip(grads, decoder.trainable_variables), optimizer.apply_gradients(zip(grads, decoder.trainable_variables),
global_step=global_step_decoder) global_step=global_step_decoder)
@ -535,8 +539,10 @@ def _train_zdiscriminator_step(z_discriminator: ZDiscriminator, encoder: Encoder
step=global_step) step=global_step)
summary_ops_v2.scalar(name='z_discriminator_loss', tensor=_zd_train_loss, summary_ops_v2.scalar(name='z_discriminator_loss', tensor=_zd_train_loss,
step=global_step) step=global_step)
for variable, grad in zd_grads: for grad, variable in zip(zd_grads, z_discriminator.trainable_variables):
summary_ops_v2.histogram(name='z_discriminator_grads/' + variable.name, tensor=tf.math.l2_normalize(grad), summary_ops_v2.histogram(name='gradients/' + variable.name, tensor=tf.math.l2_normalize(grad),
step=global_step)
summary_ops_v2.histogram(name='variables/' + variable.name, tensor=tf.math.l2_normalize(variable),
step=global_step) step=global_step)
optimizer.apply_gradients(zip(zd_grads, z_discriminator.trainable_variables), optimizer.apply_gradients(zip(zd_grads, z_discriminator.trainable_variables),
global_step=global_step_zd) global_step=global_step_zd)
@ -579,8 +585,10 @@ def _train_enc_dec_step(encoder: Encoder, decoder: Decoder, z_discriminator: ZDi
step=global_step) step=global_step)
summary_ops_v2.scalar(name='encoder_decoder_loss', tensor=_enc_dec_train_loss, summary_ops_v2.scalar(name='encoder_decoder_loss', tensor=_enc_dec_train_loss,
step=global_step) step=global_step)
for variable, grad in enc_dec_grads: for grad, variable in zip(enc_dec_grads, encoder.trainable_variables + decoder.trainable_variables):
summary_ops_v2.histogram(name='encoder_decoder_grads/' + variable.name, tensor=tf.math.l2_normalize(grad), summary_ops_v2.histogram(name='gradients/' + variable.name, tensor=tf.math.l2_normalize(grad),
step=global_step)
summary_ops_v2.histogram(name='variables/' + variable.name, tensor=tf.math.l2_normalize(variable),
step=global_step) step=global_step)
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),