Learning Decay Example

      self.tensor_objects['Optimizer'] = self.optimizer()

    elif obj == 'train_op':

      self.tensor_objects['train_op'] = self._get_tf('Optimizer').minimize(

          self.loss.out_tensor)

          self.loss.out_tensor, global_step=self._get_tf('GlobalStep'))

    elif obj == 'summary_op':

      self.tensor_objects['summary_op'] = tf.summary.merge_all(

          key=tf.GraphKeys.SUMMARIES)

    elif obj == 'GlobalStep':

      with self._get_tf("Graph").as_default():

        self.tensor_objects['GlobalStep'] = tf.Variable(0, trainable=False)

    return self._get_tf(obj)

  def _initialize_weights(self, sess, saver):

import numpy as np

from models.tensorgraph import TFWrapper

np.random.seed(123)

import tensorflow as tf

model = GraphConvTensorGraph(

    len(tox21_tasks), batch_size=batch_size, mode='classification')

global_step = model._get_tf('GlobalStep')

def optimizer_function():

    starter_learning_rate = 0.1

    learning_rate = tf.train.exponential_decay(starter_learning_rate, global_step,

                                               100000, 0.96, staircase=True)

    return tf.train.GradientDescentOptimizer(learning_rate)

model.set_optimizer(TFWrapper(optimizer_function))

model.fit(train_dataset, nb_epoch=10)

print("Evaluating model")