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

Added logging of loss values as summaries

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-02-08 06:51:40 +01:00
parent 65638974d3
commit a6e90e70b3
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/twomartens/masterthesis/aae/train.py
View File

@ -143,6 +143,7 @@ def train_mnist(folding_id: int, inlier_classes: Sequence[int], total_classes: i
print("learning rate change!")
nr_batches = len(mnist_train_x) // batch_size
log_frequency = 10
for it in range(nr_batches):
x = k.expand_dims(extract_batch(mnist_train_x, it, batch_size))
# x discriminator
@ -189,6 +190,14 @@ def train_mnist(folding_id: int, inlier_classes: Sequence[int], total_classes: i
global_step=global_step_enc_dec)
enc_dec_loss_avg(reconstruction_loss)
encoder_loss_avg(encoder_loss)
if it % log_frequency == 0:
# log the losses every log frequency batches
summary_ops_v2.scalar('encoder_loss', encoder_loss_avg.result(), step=global_step_enc_dec)
summary_ops_v2.scalar('decoder_loss', decoder_loss_avg.result(), step=global_step_decoder)
summary_ops_v2.scalar('encoder_decoder_loss', enc_dec_loss_avg.result(), step=global_step_enc_dec)
summary_ops_v2.scalar('z_discriminator_loss', zd_loss_avg.result(), step=global_step_zd)
summary_ops_v2.scalar('x_discriminator_loss', xd_loss_avg.result(), step=global_step_xd)
if it == 0:
directory = 'results' + str(inlier_classes[0])