def run_case(data, checkpoint_dir, use_gateway=True):
# reshape data to TF format
data = np.transpose(data, (0, 2, 3, 1))
# reset the Graph
tf.reset_default_graph()
# tf Graph input
x = tf.placeholder(tf.float32, shape=data.shape)
# Construct model
pred = concat_conv_net(x, use_gateway)
# Transform to the Intel DAAL model
model = pydaal.transform(pred)
# Initializing the variables
init = tf.global_variables_initializer()
# Provide a reference path to PyDAAL model
pydaal.dump_model(model, checkpoint_dir)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=0)
# Launch the graph
with tf.Session() as sess:
sess.run(init)
checkpoint_path = os.path.join(checkpoint_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=0)
predictions = sess.run(pred, feed_dict={x: data})
return predictions
def test_reshape_case(self):
test_data = np.random.randn(100, 64, 7, 7)
reshape = tf.reshape(tf.constant(test_data), [-1, 7 * 7 * 64])
with tf.Session() as sess:
tf_predictions = sess.run(reshape)
# Transform to the Intel DAAL model
model = pydaal.transform_all()
pydaal.dump_model(model, self.checkpoint_dir)
self.net.build(self.checkpoint_dir)
with self.net.predict(test_data) as predictions:
assert_allclose(tf_predictions, predictions, atol=1e-10)
def run_case(data, checkpoint_dir):
# reset the Graph
tf.reset_default_graph()
# tf Graph input
x = tf.placeholder(tf.float32)
# Construct model
pred = activation_net(x)
# Launch the graph
with tf.Session() as sess:
predictions = sess.run(pred, feed_dict={x: data})
# Transform to the Intel DAAL model
model = pydaal.transform(pred)
# Provide a reference path to PyDAAL model
pydaal.dump_model(model, checkpoint_dir)
return predictions
def run_case(data, checkpoint_dir):
# Network Parameters
n_classes = 10
# reset the Graph
tf.reset_default_graph()
# tf Graph input
x = tf.placeholder(tf.float32)
dropout = tf.placeholder(tf.float32)
# Construct model
pred = mlp_net(x, n_classes, dropout)
# Transform to the Intel DAAL model
model = pydaal.transform_all()
# Initializing the variables
init = tf.global_variables_initializer()
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=0)
# Launch the graph
with tf.Session() as sess:
sess.run(init)
checkpoint_path = os.path.join(checkpoint_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=0)
predictions = sess.run(pred, feed_dict={x: data, dropout: 1.})
# Provide a reference path to PyDAAL model
pydaal.dump_model(model, checkpoint_dir)
return predictions