cifar_basic2.py

def get_batch(x, y, batch_size):
#    print("x shape : {0}".format(x.shape))
#    print("y shape : {0}".format(y.shape))
    n_samples = x.shape[0]
    indices = np.random.choice(n_samples, batch_size)
    return x[indices], y[indices], indices

# clear old variables
tf.reset_default_graph()
X = tf.placeholder(tf.float32, [None, 32, 32, 3], name="Input")
y = tf.placeholder(tf.int64, [None])

# Placeholders for batchnorm and dropout
is_training = tf.placeholder(tf.bool)
keep_prob = tf.placeholder(tf.float32)

# setup variables
Wconv1 = tf.get_variable("Wconv1", shape=[7, 7, 3, 32])
bconv1 = tf.get_variable("bconv1", shape=[32])
W1 = tf.get_variable("W1", shape=[5408, 10])
b1 = tf.get_variable("b1", shape=[10])

# define our graph (e.g. two_layer_convnet)
a1 = tf.nn.conv2d(X, Wconv1, strides=[1,2,2,1], padding='VALID') + bconv1
h1 = tf.nn.relu(a1, name="OutputConvRelu_1")
h1_flat = tf.reshape(h1,[-1,5408])
y_out = tf.matmul(h1_flat,W1,name="Output") + b1

# define our loss
total_loss = tf.losses.hinge_loss(tf.one_hot(y,10),logits=y_out)
mean_loss = tf.reduce_mean(total_loss)
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=tf.one_hot(y, depth=10), logits=y_out))

# define our optimizer
optimizer = tf.train.AdamOptimizer(5e-4) # select optimizer and set learning rate
train_step = optimizer.minimize(mean_loss)

correct_prediction = tf.equal(tf.argmax(y_out, 1), y)
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

loss_history = []
train_accuracy_history = []
val_accuracy_history = []

# Add ops to save and restore all the variables.
saver = tf.train.Saver()

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for i in range(10000):
        batch = get_batch(X_train, y_train, 50)
        if i % 400 == 0:
            train_accuracy = accuracy.eval(feed_dict={X: batch[0], y: batch[1], keep_prob: 1.0, is_training:False})
            val_accuracy = accuracy.eval(feed_dict={X: X_val, y: y_val, keep_prob: 1.0, is_training:False})
            train_accuracy_history.append(train_accuracy)
            val_accuracy_history.append(val_accuracy)
            print('step %d, training accuracy %g, , validation accuracy %g' % (i, train_accuracy, val_accuracy))
                  
        _, loss_i = sess.run([train_step, cross_entropy], feed_dict={X: batch[0], y: batch[1], keep_prob: 0.75, is_training:True})
        loss_history.append(loss_i)

    print('test accuracy %g' % accuracy.eval(feed_dict={X: X_val, y: y_val, keep_prob: 1.0, is_training:False}))
    
    # Save the variables to disk.
    save_path = saver.save(sess, "cifar_save8/model.ckpt")
    print("Model saved in path: %s" % save_path)