def testDynamicRnnTrainingLoop(self):
graph = basic_rnn_train.make_graph(
sequence_example_file=self.sequence_example_file)
metrics = list(basic_rnn_train.training_loop(
graph, self.train_dir, num_training_steps=5, summary_frequency=1))
for metric in metrics:
self.assertTrue(metric['loss'] >= 0)
self.assertTrue(metric['accuracy'] >= 0)
def testEvalLoop(self):
train_graph = basic_rnn_train.make_graph(
sequence_example_file=self.sequence_example_file)
list(basic_rnn_train.training_loop(
train_graph, self.eval_dir, num_training_steps=5, summary_frequency=1))
eval_graph = basic_rnn_train.make_graph(
sequence_example_file=self.sequence_example_file)
metric = basic_rnn_train.eval_loop(
eval_graph, self.eval_dir, self.eval_dir,
num_training_steps=5, summary_frequency=1).next()
self.assertTrue('loss' in metric)
self.assertTrue('log_perplexity' in metric)
self.assertTrue('accuracy' in metric)