chess_deep.py

import tensorflow as tf
import sys, os, argparse
import chess_dataset

# Disable compile warnings
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

def deepnn(x):
    """deepnn builds the graph for a deep net for classifying chess pieces.

    Args:
        x: an input tensor with the dimensions (N_examples, 2500), where
        2500 is the number of pixels in a standard CHARS image.

    Returns:
        A tuple (y, keep_prob). y is a tensor of shape (N_examples, 7), with values
        equal to the logits of classifying the piece into one of 7 classes.
        keep_prob is a scalar placeholder for the probability of
        dropout.
    """

    # First convolutional layer - maps one rgb image to 32 feature maps.
    W_conv1 = weight_variable([5, 5, 3, 32])
    b_conv1 = bias_variable([32])
    h_conv1 = tf.nn.relu(conv2d(x, W_conv1) + b_conv1)

    # Pooling layer - downsamples by 2X.
    h_pool1 = max_pool_2x2(h_conv1)

    # Second convolutional layer -- maps 32 feature maps to 64.
    W_conv2 = weight_variable([5, 5, 32, 64])
    b_conv2 = bias_variable([64])
    h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)

    # Second pooling layer.
    h_pool2 = max_pool_2x2(h_conv2)

    # Fully connected layer 1 -- after 2 round of downsampling, our 20x20 image
    # is down to 5x5x64 feature maps -- maps this to 1024 features.
    W_fc1 = weight_variable([13 * 13 * 64, 1024])
    b_fc1 = bias_variable([1024])

    h_pool2_flat = tf.reshape(h_pool2, [-1, 13*13*64])
    h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

    # Dropout - controls the complexity of the model, prevents co-adaptation of
    # features.
    keep_prob = tf.placeholder(tf.float32)
    h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

    # Map the 1024 features to 7 classes, one for each piece
    W_fc2_p = weight_variable([1024, 7])
    b_fc2_p = bias_variable([7])

    # Map the 1024 features to 3 classes, one for each color (white, black, none)
    W_fc2_c = weight_variable([1024, 3])
    b_fc2_c = bias_variable([3])

    y_conv_p = tf.matmul(h_fc1_drop, W_fc2_p) + b_fc2_p
    y_conv_c = tf.matmul(h_fc1_drop, W_fc2_c) + b_fc2_c
    return y_conv_p, y_conv_c, keep_prob


def conv2d(x, W):
    """conv2d returns a 2d convolution layer with full stride."""
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')


def max_pool_2x2(x):
    """max_pool_2x2 downsamples a feature map by 2X."""
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
                          strides=[1, 2, 2, 1], padding='SAME')


def weight_variable(shape):
    """weight_variable generates a weight variable of a given shape."""
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)


def bias_variable(shape):
    """bias_variable generates a bias variable of a given shape."""
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)


def main():
    # Import data
    chess = chess_dataset.read_data_sets()

    # Create the model
    x = tf.placeholder(tf.float32, [None, 50, 50, 3])

    # Define loss and optimizer
    p_y_ = tf.placeholder(tf.float32, [None, 7])
    c_y_ = tf.placeholder(tf.float32, [None, 3])

    # Build the graph for the deep net
    y_conv_p, y_conv_c, keep_prob = deepnn(x)

    cross_entropy_p = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(labels=p_y_, logits=y_conv_p))
    cross_entropy_c = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(labels=c_y_, logits=y_conv_c))
    joint_loss = cross_entropy_p + cross_entropy_c
    train_step = tf.train.AdamOptimizer(1e-4).minimize(joint_loss)
    correct_prediction_p = tf.equal(tf.argmax(y_conv_p, 1), tf.argmax(p_y_, 1))
    correct_prediction_c = tf.equal(tf.argmax(y_conv_c, 1), tf.argmax(c_y_, 1))
    accuracy_p = tf.reduce_mean(tf.cast(correct_prediction_p, tf.float32))
    accuracy_c = tf.reduce_mean(tf.cast(correct_prediction_c, tf.float32))

    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for i in range(20000):
            batch = chess.train.next_batch(50)
            if i % 100 == 0:
                train_accuracy_p = accuracy_p.eval(feed_dict={
                    x: batch[0], p_y_: batch[1], keep_prob: 1.0})
                train_accuracy_c = accuracy_c.eval(feed_dict={
                    x: batch[0], c_y_: batch[2], keep_prob: 1.0})
                test_accuracy_p = accuracy_p.eval(feed_dict={
                    x: chess.test.images, p_y_: chess.test.p_labels, keep_prob: 1.0})
                test_accuracy_c = accuracy_c.eval(feed_dict={
                    x: chess.test.images, c_y_: chess.test.c_labels, keep_prob: 1.0})
                print('step {} accuracy: train p {:.2f}, train c {:.2f}'.format(
                    i, train_accuracy_p, train_accuracy_c))
                print('                  test  p {:.2f}, test  c {:.2f}'.format(
                    test_accuracy_p, test_accuracy_c))
            train_step.run(feed_dict={x: batch[0], p_y_: batch[1], c_y_: batch[2], keep_prob: 0.5})

if __name__ == '__main__':
    main()