def show_maps():
layer = model.h_conv
data = Data(1)
# print data._y.shape
# print data._X.shape
# layer = model.h_conv5
batch_size = 40 # image that will be used, taken from test_data/X.npy
batch = data.next_batch(batch_size) # jump batch_size images
batch = data.next_batch(batch_size) # use the first image of this batch
feature_maps = sess.run(layer,
feed_dict={
model.x: batch[0],
model.y_: batch[1],
model.keep_prob: 1.0
})
plotFeatureMaps(feature_maps)
loss = tf.reduce_mean(tf.square(tf.subtract(
model.y_, model.y))) + tf.add_n([tf.nn.l2_loss(v)
for v in train_vars]) * L2NormConst
train_step = tf.train.AdamOptimizer(1e-4).minimize(loss)
sess.run(tf.global_variables_initializer())
# Training loop variables
epochs = 100
batch_size = 50
num_samples = data.num_examples
step_size = int(num_samples / batch_size)
for epoch in range(epochs):
for i in range(step_size):
batch = data.next_batch(batch_size)
train_step.run(feed_dict={
model.x: batch[0],
model.y_: batch[1],
model.keep_prob: 0.8
})
if i % 10 == 0:
loss_value = loss.eval(feed_dict={
model.x: batch[0],
model.y_: batch[1],
model.keep_prob: 1.0
})
print("epoch: %d step: %d loss: %g" %
(epoch, epoch * batch_size + i, loss_value))
sess.run(tf.global_variables_initializer())
# Training loop variables
epochs = 30
batch_size = 64
num_train_samples = train_data.num_examples # train ~6500 test 520
num_train_batches = int(num_train_samples / batch_size) # train 130, test 11
step_size = int(num_train_samples / batch_size) # 11
train_loss_values = []
test_loss_values = []
for epoch in range(epochs):
for i in range(num_train_batches):
count = epoch * num_train_batches + i # batches seen
train_batch = train_data.next_batch(batch_size)
test_batch = test_data.next_batch(batch_size)
train_step.run(
feed_dict={
model.x: train_batch[0],
model.y_: train_batch[1],
model.keep_prob: 0.7
})
test_loss_value = test_loss.eval(feed_dict={
model.x: test_batch[0],
model.y_: test_batch[1],
model.keep_prob: 1.0
})
if count % 100 == 0:
train_loss_value = loss.eval(
feed_dict={
