def evaluate():
"""Run Eval once.
"""
# get a list of image filenames
filenames = glb('/data/fluid_flow_steady_state_128x128_test/*')
filenames.sort(key=alphanum_key)
filename_len = len(filenames)
shape = [128, 256]
with tf.Graph().as_default():
# Make image placeholder
boundary_op = tf.placeholder(tf.float32, [1, shape[0], shape[1], 1])
# Build a Graph that computes the logits predictions from the
# inference model.
sflow_p = flow_net.inference(boundary_op, 1.0)
# Restore the moving average version of the learned variables for eval.
variables_to_restore = tf.all_variables()
saver = tf.train.Saver(variables_to_restore)
sess = tf.Session()
ckpt = tf.train.get_checkpoint_state(TEST_DIR)
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = 1
graph_def = tf.get_default_graph().as_graph_def(add_shapes=True)
for run in filenames:
# read in boundary
flow_name = run + '/fluid_flow_0002.h5'
boundary_np = load_boundary(flow_name, shape).reshape(
[1, shape[0], shape[1], 1])
sflow_true = load_flow(flow_name, shape)
# calc logits
sflow_generated = sess.run(sflow_p,
feed_dict={boundary_op: boundary_np})[0]
if FLAGS.display_test:
# convert to display
sflow_plot = np.concatenate([
sflow_true, sflow_generated, sflow_true - sflow_generated
],
axis=1)
boundary_concat = np.concatenate(3 * [boundary_np], axis=2)
sflow_plot = np.sqrt(
np.square(sflow_plot[:, :, 0]) +
np.square(sflow_plot[:, :, 1])
) - .05 * boundary_concat[0, :, :, 0]
# display it
plt.imshow(sflow_plot)
plt.colorbar()
plt.show()
print("the percent error on " + FLAGS.test_set + " is")
print(p_error)
def evaluate():
"""Run Eval once.
Args:
saver: Saver.
summary_writer: Summary writer.
top_k_op: Top K op.
summary_op: Summary op.
"""
shape = [128, 256]
with tf.Graph().as_default():
# Make image placeholder
boundary_op = tf.placeholder(tf.float32, [8, shape[0], shape[1], 1])
# Build a Graph that computes the logits predictions from the
# inference model.
sflow_p = flow_net.inference(boundary_op, 1.0)
# Restore the moving average version of the learned variables for eval.
variables_to_restore = tf.all_variables()
saver = tf.train.Saver(variables_to_restore)
sess = tf.Session()
ckpt = tf.train.get_checkpoint_state(TEST_DIR)
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = 1
graph_def = tf.get_default_graph().as_graph_def(add_shapes=True)
num_runs = 1000
boundary_np = np.zeros([8, shape[0], shape[1], 1])
_ = sess.run(sflow_p, feed_dict={boundary_op: boundary_np})[0]
t = time.time()
for i in xrange(num_runs):
# make boundary
# calc logits
_ = sess.run(sflow_p, feed_dict={boundary_op: boundary_np})[0]
elapsed = time.time() - t
print("time per inpu is ")
print(elapsed / (num_runs * 8.))
def train():
"""Train ring_net for a number of steps."""
with tf.Graph().as_default():
# make inputs
boundary, sflow = flow_net.inputs(FLAGS.batch_size)
# create and unrap network
sflow_p = flow_net.inference(boundary, FLAGS.keep_prob)
# calc error
error = flow_net.loss_image(sflow_p, sflow)
# train hopefuly
train_op = flow_net.train(error, FLAGS.learning_rate)
# List of all Variables
variables = tf.global_variables()
# Build a saver
saver = tf.train.Saver(tf.global_variables())
#for i, variable in enumerate(variables):
# print '----------------------------------------------'
# print variable.name[:variable.name.index(':')]
# Summary op
summary_op = tf.summary.merge_all()
# Build an initialization operation to run below.
init = tf.global_variables_initializer()
# Start running operations on the Graph.
sess = tf.Session()
# init if this is the very time training
sess.run(init)
# init from checkpoint
saver_restore = tf.train.Saver(variables)
ckpt = tf.train.get_checkpoint_state(TRAIN_DIR)
if ckpt is not None:
print("init from " + TRAIN_DIR)
try:
saver_restore.restore(sess, ckpt.model_checkpoint_path)
except:
tf.gfile.DeleteRecursively(TRAIN_DIR)
tf.gfile.MakeDirs(TRAIN_DIR)
print("there was a problem using variables in checkpoint, random init will be used instead")
# Start que runner
tf.train.start_queue_runners(sess=sess)
# Summary op
graph_def = sess.graph.as_graph_def(add_shapes=True)
summary_writer = tf.summary.FileWriter(TRAIN_DIR, graph_def=graph_def)
for step in range(FLAGS.max_steps):
t = time.time()
_ , loss_value = sess.run([train_op, error],feed_dict={})
elapsed = time.time() - t
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step%100 == 0:
summary_str = sess.run(summary_op, feed_dict={})
summary_writer.add_summary(summary_str, step)
print("loss value at " + str(loss_value))
print("time per batch is " + str(elapsed))
if step%1000 == 0:
checkpoint_path = os.path.join(TRAIN_DIR, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
print("saved to " + TRAIN_DIR)
