# gpu_opinions = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
with tf.Session(config=tf_config) as sess:
saver = tf.train.Saver(max_to_keep=None)
# Initialize parameters
# sess.run(tf.global_variables_initializer())
saver.restore(sess, dir_load)
#Output directory for models and summaries
timestamp = str(int(time.time()))
out_dir = os.path.abspath(os.path.join(os.path.curdir, "run", timestamp))
print("Writing to {}\n".format(out_dir))
#testing
test_batches = data_process.batch_iter1(test_file, test_batchSize)
testOutAll = []
count = 0
for batch in test_batches:
count += 1
testLoss_out, test_out = sess.run([loss, y_conv],
feed_dict={
x: batch[0],
y_: batch[1],
keep_prob: 1.0
})
print("step %d, test loss %g" % (count * test_batchSize, testLoss_out))
testOut_flat = test_out.reshape(-1)
testOutAll = np.concatenate([testOutAll, testOut_flat])
# print("the predict:", test_out)
def main(_):
#Graph
with tf.name_scope('input'):
x = tf.placeholder(tf.float32,
[None, 15 * 4 * 101 * 101]) # input data
y_ = tf.placeholder(tf.float32, [
None,
]) # label
keep_prob = tf.placeholder(tf.float32)
with tf.name_scope('input_reshape'):
x_image = tf.reshape(x, [-1, 4, 101, 101, 1])
# x_image = tf.transpose(x_image, [0,3,4,1,2])
x_image_h0 = x_image[:, 0, :, :, :]
x_image_h1 = x_image[:, 1, :, :, :]
x_image_h2 = x_image[:, 2, :, :, :]
x_image_h3 = x_image[:, 3, :, :, :]
print(x_image_h1.shape)
fc1_h0 = net(x_image_h0, 0, keep_prob)
fc1_h1 = net(x_image_h1, 1, keep_prob)
fc1_h2 = net(x_image_h2, 2, keep_prob)
fc1_h3 = net(x_image_h3, 3, keep_prob)
fc1_all = tf.concat([fc1_h0, fc1_h1, fc1_h2, fc1_h3], axis=1)
print(fc1_all.shape)
fc2 = slim.fully_connected(inputs=fc1_all,
num_outputs=20,
activation_fn=None,
scope='fc2')
y_conv = slim.fully_connected(inputs=fc2,
num_outputs=1,
activation_fn=None,
scope='fc3')
y_conv = tf.reshape(y_conv, [-1])
# loss function
loss = tf.sqrt(tf.reduce_mean(tf.square(tf.subtract(y_conv, y_))))
train_step = tf.train.AdamOptimizer(1e-4).minimize(loss)
# gpu configuration
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
tf_config.gpu_options.visible_device_list = '0'
# gpu_opinions = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
with tf.Session(config=tf_config) as sess:
saver = tf.train.Saver(max_to_keep=None)
# Initialize parameters
sess.run(tf.global_variables_initializer())
# saver.restore(sess, dir_load)
#Output directory for models and summaries
timestamp = str(int(time.time()))
out_dir = os.path.abspath(
os.path.join(os.path.curdir, "run", timestamp))
print("Writing to {}\n".format(out_dir))
#summary for loss
tf.summary.scalar("loss", loss)
tf.summary.histogram("fc1_all", fc1_all)
# tf.summary.scalar("learning rate", learning_rate)
# loss_summary = tf.summary.scalar("learning rate", learning_rate)
# tf.summary.image("input", x_image_h1, max_outputs = 15)
merged = tf.summary.merge_all() # it is useful
# grad_summaries = []
# grad_summaries.append(loss_summary)
# merged = tf.summary.merge(grad_summaries)
#train summary
train_summary_dir = os.path.join(out_dir, "summaries", "train")
train_summary_writer = tf.summary.FileWriter(train_summary_dir,
sess.graph)
#dev summary
dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
dev_summary_writer = tf.summary.FileWriter(dev_summary_dir, sess.graph)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
#training
count = 0
for i in range(4):
train_batches = data_process.batch_iter1(train_file,
batch_size,
shuffle=True)
for batch in train_batches:
count += 1
# vari_x.extend(batch[2])
# vari_y.extend(batch[3])
#training with a batch
sess.run(train_step,
feed_dict={
x: batch[0],
y_: batch[1],
keep_prob: 0.6
})
#training loss
train_summary, train_loss_out = sess.run([merged, loss],
feed_dict={
x: batch[0],
y_: batch[1],
keep_prob: 1
})
print("epochs %d, step %d, training loss %g" %
(i, count * batch_size, train_loss_out))
train_summary_writer.add_summary(train_summary, count)
#varification
# if (count * batch_size) % 3000 == 0:
if count % 30 == 0:
veri_batches = data_process.batch_iter1(
veri_file, veri_size)
testOutAll = []
y_all = []
for veri_batch in veri_batches:
vari_summary, veri_loss_out, y_veri = sess.run(
[merged, loss, y_conv],
feed_dict={
x: veri_batch[0],
y_: veri_batch[1],
keep_prob: 1
})
y_all = np.concatenate([y_all, veri_batch[1]])
testOutAll = np.concatenate([testOutAll, y_veri])
# print(y_all.shape, testOutAll.shape)
err_2 = (testOutAll - y_all)**2
rmse = math.sqrt(np.mean(err_2))
print("*******epochs %d, step %d, varification loss %g" %
(i, count * batch_size, rmse))
dev_summary_writer.add_summary(vari_summary, count)
#save model
# if (count * batch_size) % 3000 == 0:
if count % 30 == 0:
# if not os.path.exists(dir_model):
path = saver.save(sess,
checkpoint_prefix,
global_step=count * batch_size)
print("Saved model checkpoint to {}\n".format(path))
train_summary_dir = os.path.join(out_dir, "summaries", "train")
train_summary_writer = tf.summary.FileWriter(train_summary_dir, sess.graph)
#dev summary
dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
dev_summary_writer = tf.summary.FileWriter(dev_summary_dir, sess.graph)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
#training
count = 0
for i in range(4):
train_batches = data_process.batch_iter1(train_file, batch_size)
for batch in train_batches:
count += 1
# vari_x.extend(batch[2])
# vari_y.extend(batch[3])
#training with a batch
sess.run(train_step,
feed_dict={
x: batch[0],
y_: batch[1],
keep_prob: 0.6
})
#training loss
train_summary, train_loss_out = sess.run([merged, loss],
feed_dict={
x: batch[0],