pre_ind.append(i)
post_ind.append(n)
weights.append(ip_weights[i, n])
def create_ipb_connections(ip_n, ip_bias, pre_ind, post_ind, weights):
for i in range(ip_n):
pre_ind.append(0)
post_ind.append(i)
weights.append(ip_bias[i])
#------------------------------------------------------------------------------
# load CIFAR10
#------------------------------------------------------------------------------
cifar10_data_set = cifar10_extract.Cifar10DataSet('../dataset/')
test_images, test_labels = cifar10_data_set.test_data()
# crop
test_images = test_images[:, 4:28, 4:28, :] * 255
#------------------------------------------------------------------------------
# set parameters and equations
#------------------------------------------------------------------------------
eqs_input = '''rates : Hz
dv/dt = 1 : second'''
input_thresh = 'v > 1/rates'
input_reset = 'v -= 1/rates'
eqs_conv1 = 'v : 1'
eqs_pool1 = 'v : 1'
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get images and labels for CIFAR-10.
# eval_data = FLAGS.eval_data == 'test'
# images, labels = cifar10.inputs(eval_data=eval_data)
images = tf.placeholder('float', [FLAGS.batch_size, 24, 24, 3])
labels = tf.placeholder('int32', [FLAGS.batch_size, 10])
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# lp = tf.argmax(logits, 1)
# count correct prediction
correct_prediction = tf.equal(tf.argmax(logits, 1),
tf.argmax(labels, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
# Calculate predictions.
# top_k_op = tf.nn.in_top_k(logits, labels, 1)
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
# eval_once(saver, summary_writer, top_k_op, summary_op)
# get cifar-10 dataset
cifar10_data_set = cifar10_extract.Cifar10DataSet('../dataset/')
test_images, test_labels = cifar10_data_set.test_data()
# crop
test_images = test_images[:, 4:28, 4:28, :] - 0.5
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
# Restores from checkpoint
saver.restore(sess, ckpt.model_checkpoint_path)
else:
print('No checkpoint file found')
return
# avg = 0
# for i in xrange(100):
# batch_xs = test_images[
# i * FLAGS.batch_size:(i + 1) * FLAGS.batch_size]
# batch_ys = test_labels[
# i * FLAGS.batch_size:(i + 1) * FLAGS.batch_size]
# avg += sess.run(accuracy, feed_dict={images:
# batch_xs, labels: batch_ys})
# avg /= (100)
# print("test accuracy %g" % avg)
# Get conv1 output
conv1 = tf.get_default_graph().get_tensor_by_name('conv1/conv1:0')
cnn_conv1 = sess.run(conv1,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_conv1 = cnn_conv1.reshape((FLAGS.batch_size, 24 * 24 * 64))
# Get pool1 output
pool1 = tf.get_default_graph().get_tensor_by_name('pool1:0')
cnn_pool1 = sess.run(pool1,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_pool1 = cnn_pool1.reshape((FLAGS.batch_size, 12 * 12 * 64))
# Get conv2 output
conv2 = tf.get_default_graph().get_tensor_by_name('conv2/conv2:0')
cnn_conv2 = sess.run(conv2,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_conv2 = cnn_conv2.reshape((FLAGS.batch_size, 12 * 12 * 64))
# Get pool2 output
pool2 = tf.get_default_graph().get_tensor_by_name('pool2:0')
cnn_pool2 = sess.run(pool2,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_pool2 = cnn_pool2.reshape((FLAGS.batch_size, 6 * 6 * 64))
# Get local3 output
local3 = tf.get_default_graph().get_tensor_by_name(
'local3/local3:0')
cnn_local3 = sess.run(local3,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_local3 = cnn_local3.reshape((FLAGS.batch_size, 384))
# Get local4 output
local4 = tf.get_default_graph().get_tensor_by_name(
'local4/local4:0')
cnn_local4 = sess.run(local4,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_local4 = cnn_local4.reshape((FLAGS.batch_size, 192))
# Get softmax_linear output
local5 = tf.get_default_graph().get_tensor_by_name(
'softmax_linear/softmax_linear:0')
cnn_local5 = sess.run(local5,
feed_dict={
images: test_images[0:FLAGS.batch_size],
labels: test_labels[0:FLAGS.batch_size]
})
cnn_local5 = cnn_local5.reshape((FLAGS.batch_size, 10))
sio.savemat(
'output/cifar10_cnn.mat', {
'conv1': cnn_conv1,
'pool1': cnn_pool1,
'conv2': cnn_conv2,
'pool2': cnn_pool2,
'ip1': cnn_local3,
'ip2': cnn_local4,
'ip3': cnn_local5
})
# Save weights for brian2
with tf.variable_scope('conv1', reuse=True):
w_conv1 = tf.get_variable("weights")
b_conv1 = tf.get_variable("biases")
with tf.variable_scope('conv2', reuse=True):
w_conv2 = tf.get_variable("weights")
b_conv2 = tf.get_variable("biases")
with tf.variable_scope('local3', reuse=True):
w_local3 = tf.get_variable("weights")
b_local3 = tf.get_variable("biases")
with tf.variable_scope('local4', reuse=True):
w_local4 = tf.get_variable("weights")
b_local4 = tf.get_variable("biases")
with tf.variable_scope('softmax_linear', reuse=True):
w_local5 = tf.get_variable("weights")
b_local5 = tf.get_variable("biases")
if not tf.gfile.Exists('output/weights'):
tf.gfile.MakeDirs('output/weights')
save_w_conv1 = sess.run(w_conv1)
save_b_conv1 = sess.run(b_conv1)
save_w_conv2 = sess.run(w_conv2)
save_b_conv2 = sess.run(b_conv2)
save_w_local3 = sess.run(w_local3)
save_b_local3 = sess.run(b_local3)
save_w_local4 = sess.run(w_local4)
save_b_local4 = sess.run(b_local4)
save_w_local5 = sess.run(w_local5)
save_b_local5 = sess.run(b_local5)
sio.savemat(
'output/weights/cifar10_weights.mat', {
'conv1_w': save_w_conv1,
'conv1_b': save_b_conv1,
'conv2_w': save_w_conv2,
'conv2_b': save_b_conv2,
'ip1_w': save_w_local3,
'ip1_b': save_b_local3,
'ip2_w': save_w_local4,
'ip2_b': save_b_local4,
'ip3_w': save_w_local5,
'ip3_b': save_b_local5
})