Problem GAN conversion when applying variable reuse on tensorflow











up vote
1
down vote

favorite












I am building an GAN and when i started calling my discriminator twice, using reuse, my GAN started to diverge. I first created my discriminator as following:



def discriminator(self, x_past, x_future, gen_future):
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
with tf.variable_scope("disc") as disc:
gen_future = tf.concat([gen_future, x_past], 2)
x_future = tf.concat([x_future, x_past], 2)
x_in = tf.concat([gen_future, x_future], 0)
conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
padding='same', activation=tf.nn.relu)
max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
padding='same', activation=tf.nn.relu)
max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

# Flatten and add dropout
flat = tf.reshape(max_pool_2, (-1, 9))
flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

# Predictions
logits = tf.layers.dense(flat, 2)

y_true = logits[:self.batch_size]
y_gen = logits[self.batch_size:]

return y_true, y_gen


And I was calling it like this:



y_true, y_gen = self.discriminator(x_past, x_future, gen_future)


I was able to train the GAN properly. Now I need to use reuse to be able to call it without having to send real and fake data at once. I changed it to:



def discriminator(self, x_past, x_future, reuse=False):
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
with tf.variable_scope("disc", reuse=reuse) as disc:
x_in = tf.concat([x_future, x_past], 2)
conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
padding='same', activation=tf.nn.relu)
max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
padding='same', activation=tf.nn.relu)
max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

# Flatten and add dropout
flat = tf.reshape(max_pool_2, (-1, 9))
flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

# Predictions
logits = tf.layers.dense(flat, 2)
return logits


And calling it like this:



y_true = self.discriminator(x_past, x_future)
y_gen = self.discriminator(x_past, gen_future, reuse=True)


Now it started to diverge. Any idea why is that?










share|improve this question


























    up vote
    1
    down vote

    favorite












    I am building an GAN and when i started calling my discriminator twice, using reuse, my GAN started to diverge. I first created my discriminator as following:



    def discriminator(self, x_past, x_future, gen_future):
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
    with tf.variable_scope("disc") as disc:
    gen_future = tf.concat([gen_future, x_past], 2)
    x_future = tf.concat([x_future, x_past], 2)
    x_in = tf.concat([gen_future, x_future], 0)
    conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
    padding='same', activation=tf.nn.relu)
    max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
    conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
    padding='same', activation=tf.nn.relu)
    max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

    # Flatten and add dropout
    flat = tf.reshape(max_pool_2, (-1, 9))
    flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

    # Predictions
    logits = tf.layers.dense(flat, 2)

    y_true = logits[:self.batch_size]
    y_gen = logits[self.batch_size:]

    return y_true, y_gen


    And I was calling it like this:



    y_true, y_gen = self.discriminator(x_past, x_future, gen_future)


    I was able to train the GAN properly. Now I need to use reuse to be able to call it without having to send real and fake data at once. I changed it to:



    def discriminator(self, x_past, x_future, reuse=False):
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
    with tf.variable_scope("disc", reuse=reuse) as disc:
    x_in = tf.concat([x_future, x_past], 2)
    conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
    padding='same', activation=tf.nn.relu)
    max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
    conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
    padding='same', activation=tf.nn.relu)
    max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

    # Flatten and add dropout
    flat = tf.reshape(max_pool_2, (-1, 9))
    flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

    # Predictions
    logits = tf.layers.dense(flat, 2)
    return logits


    And calling it like this:



    y_true = self.discriminator(x_past, x_future)
    y_gen = self.discriminator(x_past, gen_future, reuse=True)


    Now it started to diverge. Any idea why is that?










    share|improve this question
























      up vote
      1
      down vote

      favorite









      up vote
      1
      down vote

      favorite











      I am building an GAN and when i started calling my discriminator twice, using reuse, my GAN started to diverge. I first created my discriminator as following:



      def discriminator(self, x_past, x_future, gen_future):
      os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
      with tf.variable_scope("disc") as disc:
      gen_future = tf.concat([gen_future, x_past], 2)
      x_future = tf.concat([x_future, x_past], 2)
      x_in = tf.concat([gen_future, x_future], 0)
      conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
      conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

      # Flatten and add dropout
      flat = tf.reshape(max_pool_2, (-1, 9))
      flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

      # Predictions
      logits = tf.layers.dense(flat, 2)

      y_true = logits[:self.batch_size]
      y_gen = logits[self.batch_size:]

      return y_true, y_gen


      And I was calling it like this:



      y_true, y_gen = self.discriminator(x_past, x_future, gen_future)


      I was able to train the GAN properly. Now I need to use reuse to be able to call it without having to send real and fake data at once. I changed it to:



      def discriminator(self, x_past, x_future, reuse=False):
      os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
      with tf.variable_scope("disc", reuse=reuse) as disc:
      x_in = tf.concat([x_future, x_past], 2)
      conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
      conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

      # Flatten and add dropout
      flat = tf.reshape(max_pool_2, (-1, 9))
      flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

      # Predictions
      logits = tf.layers.dense(flat, 2)
      return logits


      And calling it like this:



      y_true = self.discriminator(x_past, x_future)
      y_gen = self.discriminator(x_past, gen_future, reuse=True)


      Now it started to diverge. Any idea why is that?










      share|improve this question













      I am building an GAN and when i started calling my discriminator twice, using reuse, my GAN started to diverge. I first created my discriminator as following:



      def discriminator(self, x_past, x_future, gen_future):
      os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
      with tf.variable_scope("disc") as disc:
      gen_future = tf.concat([gen_future, x_past], 2)
      x_future = tf.concat([x_future, x_past], 2)
      x_in = tf.concat([gen_future, x_future], 0)
      conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
      conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

      # Flatten and add dropout
      flat = tf.reshape(max_pool_2, (-1, 9))
      flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

      # Predictions
      logits = tf.layers.dense(flat, 2)

      y_true = logits[:self.batch_size]
      y_gen = logits[self.batch_size:]

      return y_true, y_gen


      And I was calling it like this:



      y_true, y_gen = self.discriminator(x_past, x_future, gen_future)


      I was able to train the GAN properly. Now I need to use reuse to be able to call it without having to send real and fake data at once. I changed it to:



      def discriminator(self, x_past, x_future, reuse=False):
      os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
      with tf.variable_scope("disc", reuse=reuse) as disc:
      x_in = tf.concat([x_future, x_past], 2)
      conv1 = tf.layers.conv1d(inputs=x_in, filters=20, kernel_size=3, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_1 = tf.layers.max_pooling1d(inputs=conv1, pool_size=2, strides=2, padding='same')
      conv2 = tf.layers.conv1d(inputs=max_pool_1, filters=3, kernel_size=2, strides=1,
      padding='same', activation=tf.nn.relu)
      max_pool_2 = tf.layers.max_pooling1d(inputs=conv2, pool_size=2, strides=2, padding='same')

      # Flatten and add dropout
      flat = tf.reshape(max_pool_2, (-1, 9))
      flat = tf.nn.dropout(flat, keep_prob=self.keep_prob)

      # Predictions
      logits = tf.layers.dense(flat, 2)
      return logits


      And calling it like this:



      y_true = self.discriminator(x_past, x_future)
      y_gen = self.discriminator(x_past, gen_future, reuse=True)


      Now it started to diverge. Any idea why is that?







      python tensorflow generative-adversarial-network






      share|improve this question













      share|improve this question











      share|improve this question




      share|improve this question










      asked Nov 10 at 22:43









      Rafael Reis

      152216




      152216





























          active

          oldest

          votes











          Your Answer






          StackExchange.ifUsing("editor", function () {
          StackExchange.using("externalEditor", function () {
          StackExchange.using("snippets", function () {
          StackExchange.snippets.init();
          });
          });
          }, "code-snippets");

          StackExchange.ready(function() {
          var channelOptions = {
          tags: "".split(" "),
          id: "1"
          };
          initTagRenderer("".split(" "), "".split(" "), channelOptions);

          StackExchange.using("externalEditor", function() {
          // Have to fire editor after snippets, if snippets enabled
          if (StackExchange.settings.snippets.snippetsEnabled) {
          StackExchange.using("snippets", function() {
          createEditor();
          });
          }
          else {
          createEditor();
          }
          });

          function createEditor() {
          StackExchange.prepareEditor({
          heartbeatType: 'answer',
          convertImagesToLinks: true,
          noModals: true,
          showLowRepImageUploadWarning: true,
          reputationToPostImages: 10,
          bindNavPrevention: true,
          postfix: "",
          imageUploader: {
          brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
          contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
          allowUrls: true
          },
          onDemand: true,
          discardSelector: ".discard-answer"
          ,immediatelyShowMarkdownHelp:true
          });


          }
          });














           

          draft saved


          draft discarded


















          StackExchange.ready(
          function () {
          StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53244152%2fproblem-gan-conversion-when-applying-variable-reuse-on-tensorflow%23new-answer', 'question_page');
          }
          );

          Post as a guest















          Required, but never shown






























          active

          oldest

          votes













          active

          oldest

          votes









          active

          oldest

          votes






          active

          oldest

          votes
















           

          draft saved


          draft discarded



















































           


          draft saved


          draft discarded














          StackExchange.ready(
          function () {
          StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53244152%2fproblem-gan-conversion-when-applying-variable-reuse-on-tensorflow%23new-answer', 'question_page');
          }
          );

          Post as a guest















          Required, but never shown





















































          Required, but never shown














          Required, but never shown












          Required, but never shown







          Required, but never shown

































          Required, but never shown














          Required, but never shown












          Required, but never shown







          Required, but never shown







          Popular posts from this blog

          Full-time equivalent

          さくらももこ

          13 indicted, 8 arrested in Calif. drug cartel investigation