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

Replaced all binary cross entropy losses with log_loss

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-02-08 21:01:38 +01:00
parent a51b535728
commit 90fa27d9ef
1 changed files with 7 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
src/twomartens/masterthesis/aae/train.py
View File

@ -50,7 +50,6 @@ from .util import prepare_image
k = tf.keras.backend k = tf.keras.backend
AdamOptimizer = tf.train.AdamOptimizer AdamOptimizer = tf.train.AdamOptimizer
tfe = tf.contrib.eager tfe = tf.contrib.eager
binary_crossentropy = tf.keras.losses.binary_crossentropy
GRACE: int = 10 GRACE: int = 10
TOTAL_LOSS_GRACE_CAP: int = 6 TOTAL_LOSS_GRACE_CAP: int = 6
@ -432,12 +431,12 @@ def _train_xdiscriminator_step(x_discriminator: XDiscriminator, decoder: Decoder
""" """
with tf.GradientTape() as tape: with tf.GradientTape() as tape:
xd_result_1 = tf.squeeze(x_discriminator(inputs)) xd_result_1 = tf.squeeze(x_discriminator(inputs))
xd_real_loss = binary_crossentropy(targets_real, xd_result_1) xd_real_loss = tf.losses.log_loss(targets_real, xd_result_1)
z = z_generator() z = z_generator()
x_fake = decoder(z) x_fake = decoder(z)
xd_result_2 = tf.squeeze(x_discriminator(x_fake)) xd_result_2 = tf.squeeze(x_discriminator(x_fake))
xd_fake_loss = binary_crossentropy(targets_fake, xd_result_2) xd_fake_loss = tf.losses.log_loss(targets_fake, xd_result_2)
_xd_train_loss = xd_real_loss + xd_fake_loss _xd_train_loss = xd_real_loss + xd_fake_loss
@ -482,7 +481,7 @@ def _train_decoder_step(decoder: Decoder, x_discriminator: XDiscriminator,
x_fake = decoder(z) x_fake = decoder(z)
xd_result = tf.squeeze(x_discriminator(x_fake)) xd_result = tf.squeeze(x_discriminator(x_fake))
_decoder_train_loss = binary_crossentropy(targets, xd_result) _decoder_train_loss = tf.losses.log_loss(targets, xd_result)
grads = tape.gradient(_decoder_train_loss, decoder.trainable_variables) grads = tape.gradient(_decoder_train_loss, decoder.trainable_variables)
if int(global_step % LOG_FREQUENCY) == 0: if int(global_step % LOG_FREQUENCY) == 0:
@ -523,11 +522,11 @@ def _train_zdiscriminator_step(z_discriminator: ZDiscriminator, encoder: Encoder
z = z_generator() z = z_generator()
zd_result = tf.squeeze(z_discriminator(z)) zd_result = tf.squeeze(z_discriminator(z))
zd_real_loss = binary_crossentropy(targets_real, zd_result) zd_real_loss = tf.losses.log_loss(targets_real, zd_result)
z = tf.squeeze(encoder(inputs)) z = tf.squeeze(encoder(inputs))
zd_result = tf.squeeze(z_discriminator(z)) zd_result = tf.squeeze(z_discriminator(z))
zd_fake_loss = binary_crossentropy(targets_fake, zd_result) zd_fake_loss = tf.losses.log_loss(targets_fake, zd_result)
_zd_train_loss = zd_real_loss + zd_fake_loss _zd_train_loss = zd_real_loss + zd_fake_loss
@ -572,8 +571,8 @@ def _train_enc_dec_step(encoder: Encoder, decoder: Decoder, z_discriminator: ZDi
x_decoded = decoder(z) x_decoded = decoder(z)
zd_result = tf.squeeze(z_discriminator(tf.squeeze(z))) zd_result = tf.squeeze(z_discriminator(tf.squeeze(z)))
encoder_loss = binary_crossentropy(targets, zd_result) * 2.0 encoder_loss = tf.losses.log_loss(targets, zd_result) * 2.0
reconstruction_loss = binary_crossentropy(inputs, x_decoded) reconstruction_loss = tf.losses.log_loss(inputs, x_decoded)
_enc_dec_train_loss = encoder_loss + reconstruction_loss _enc_dec_train_loss = encoder_loss + reconstruction_loss
enc_dec_grads = tape.gradient(_enc_dec_train_loss, enc_dec_grads = tape.gradient(_enc_dec_train_loss,