def test_add_evaluation_step(self):
with tf.Graph().as_default():
# Given
final = tf.placeholder(tf.float32, [1], name='final')
gt = tf.placeholder(tf.float32, [1], name='gt')
# When
accuracy = evaluation.evaluate(final, gt)
# Then
self.assertIsNotNone(accuracy)
self.assertEqual(type(accuracy).__name__, "Tensor")
def test_run_evaluation(self):
with tf.Graph().as_default():
with tf.Session() as sess:
# Given
final = tf.Variable([[0.1, 0.9], [0.1, 0.9]], dtype=tf.float32)
gt = tf.Variable([[0.6, 0.4], [0.2, 0.8]], dtype=tf.float32)
# When
accuracy = evaluation.evaluate(final, gt)
tf.initialize_all_variables().run()
# Then
self.assertAllClose(sess.run(accuracy), 0.5)
def main():
# Create an InteractiveSession
sess = tf.InteractiveSession()
# Remove tensorboard previous directory
if os.path.exists(FLAGS.summaries_dir):
shutil.rmtree(FLAGS.summaries_dir)
"""
Step 1 - Input data management
"""
# MNIST data
mnist = input.input_reader(FLAGS.data_dir)
# Input placeholders
x, y_ = input.input_placeholders(IMAGE_PIXELS, NUM_CLASSES)
# Reshape images for visualization
x_reshaped = tf.reshape(x, [-1, IMAGE_SIZE, IMAGE_SIZE, 1])
tf.image_summary('input', x_reshaped, NUM_CLASSES, name="y_input")
"""
Step 2 - Building the graph
"""
# Inference
softmax = graph.create_inference_step(x=x, num_pixels=IMAGE_PIXELS, num_dense1=200, num_dense2=100, num_dense3=50,
num_classes=NUM_CLASSES)
# Loss
cross_entropy = graph.add_loss_step(softmax, y_)
# Train step
train_step = graph.add_train_step(cross_entropy, FLAGS.learning_rate)
"""
Step 3 - Build the evaluation step
"""
# Model Evaluation
accuracy = evaluation.evaluate(softmax, y_)
"""
Step 4 - Merge all summaries for TensorBoard generation
"""
# Merge all the summaries and write them out to /tmp/mnist_dense_logs (by default)
merged = tf.merge_all_summaries()
train_writer = tf.train.SummaryWriter(FLAGS.summaries_dir + '/train', sess.graph)
validation_writer = tf.train.SummaryWriter(FLAGS.summaries_dir + '/validation')
"""
Step 5 - Train the model, and write summaries
"""
# Initialize all variables
tf.initialize_all_variables().run()
# All other steps, run train_step on training data, & add training summaries
for i in range(FLAGS.max_steps):
# Load next batch of data
x_batch, y_batch = mnist.train.next_batch(FLAGS.batch_size)
# Run summaries and train_step
summary, _ = sess.run([merged, train_step], feed_dict={x: x_batch, y_: y_batch})
# Add summaries to train writer
train_writer.add_summary(summary, i)
# Every 10th step, measure validation-set accuracy, and write validation summaries
if i % 10 == 0:
# Run summaries and mesure accuracy on validation set
summary, acc_valid = sess.run([merged, accuracy],
feed_dict={x: mnist.validation.images, y_: mnist.validation.labels})
# Add summaries to validation writer
validation_writer.add_summary(summary, i)
print('Validation Accuracy at step %s: %s' % (i, acc_valid))
# Measure accuracy on test set
acc_test = sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})
print('Accuracy on test set: %s' % acc_test)