from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import tensorflow as tf
# Load MNIST Data
from tensorflow.examples.tutorials.mnist import input_data
import os
import argparse

from azureml.core.run import Run

# the following 10 lines can be removed once BUG# 241943 is fixed


def get_logger():
    try:
        return Run.get_submitted_run()
    except Exception:
        return LocalLogger()


class LocalLogger:
    def log(self, key, value):
        print("AML-Log:", key, value)


def build_model(x, y_, keep_prob):
    def weight_variable(shape):
        initial = tf.truncated_normal(shape, stddev=0.1)
        return tf.Variable(initial)

    def bias_variable(shape):
        initial = tf.constant(0.1, shape=shape)
        return tf.Variable(initial)

    def conv2d(x, W):
        return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')

    def max_pool_2x2(x):
        return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')

    W_conv1 = weight_variable([5, 5, 1, 32])
    b_conv1 = bias_variable([32])

    x_image = tf.reshape(x, [-1, 28, 28, 1])

    h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
    h_pool1 = max_pool_2x2(h_conv1)

    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)
    h_pool2 = max_pool_2x2(h_conv2)

    W_fc1 = weight_variable([7 * 7 * 64, 1024])
    b_fc1 = bias_variable([1024])

    h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 64])
    h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

    h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

    W_fc2 = weight_variable([1024, 10])
    b_fc2 = bias_variable([10])

    y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2

    return y_conv


def main():
    # Get command-line arguments
    parser = argparse.ArgumentParser()
    parser.add_argument('--learning_rate', type=float,
                        default=0.0001, help='learning rate')
    parser.add_argument('--minibatch_size', type=int,
                        default=50, help='minibatch size')
    parser.add_argument('--keep_probability', type=float,
                        default=0.5, help='keep probability for dropout layer')
    parser.add_argument('--num_iterations', type=int,
                        default=1000, help='number of iterations')
    parser.add_argument('--output_dir', type=str, default='./outputs',
                        help='output directory to write checkpoints to')

    args = parser.parse_args()

    # log parameters
    run_logger = get_logger()
    run_logger.log("learning_rate", args.learning_rate)
    run_logger.log("minibatch_size", args.minibatch_size)
    run_logger.log("keep_probability", args.keep_probability)
    run_logger.log("num_iterations", args.num_iterations)

    # Load MNIST data
    mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

    sess = tf.InteractiveSession()

    x = tf.placeholder(tf.float32, shape=[None, 784])
    y_ = tf.placeholder(tf.float32, shape=[None, 10])
    keep_prob = tf.placeholder(tf.float32)

    y_conv = build_model(x, y_, keep_prob)

    cross_entropy = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))

    train_step = tf.train.AdamOptimizer(
        args.learning_rate).minimize(cross_entropy)
    correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))

    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    sess.run(tf.global_variables_initializer())

    for i in range(args.num_iterations):
        batch = mnist.train.next_batch(args.minibatch_size)
        if i % 100 == 0:
            test_acc = accuracy.eval(
                feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})
            train_accuracy = accuracy.eval(
                feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})
            print("step %d, training accuracy %g, test accuracy, %g" %
                  (i, train_accuracy, test_acc))

            # log test accuracy to AML
            run_logger.log("Accuracy", float(test_acc))
            run_logger.log("Iterations", i)

        sess.run(train_step, feed_dict={
                 x: batch[0], y_: batch[1], keep_prob: args.keep_probability})

    # Save the trained model
    model_dir = args.output_dir
    model_file = 'model.ckpt'

    os.makedirs(model_dir, exist_ok=True)

    saver = tf.train.Saver()
    saver.save(sess, os.path.join(model_dir, model_file))

    final_test_acc = sess.run(accuracy, feed_dict={
                              x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})
    run_logger.log("Accuracy", float(final_test_acc))
    run_logger.log("Iterations", args.num_iterations)
    print("test accuracy %g" % final_test_acc)


if __name__ == "__main__":
    main()