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

import os
import PIL
import PIL.Image
import numpy as np
import time
import datetime

width=28
height=28
classCount=2
channels=3

width_c2=7
height_c2=7

import tensorflow as tf
sess = tf.InteractiveSession()

def prepare():
    x=[]
    y=[]

    def add(x,y,dir,yy):
        for path in os.listdir(dir):
            xx=np.asarray(PIL.Image.open(dir+u"/"+path).convert('RGB')).reshape(width*height*channels)*1.0/255

            x += [ xx ]
            y += [ yy ]

    add(x,y,u"images28/0",[1.0,0.0])
    add(x,y,u"images28/1",[0.0,1.0])

    return np.array(x), np.array(y)

def batch_iter(data, batch_size, num_epochs, shuffle=True):
    data = np.array(data)
    data_size = len(data)
    num_batches_per_epoch = int(len(data)/batch_size) + 1
    for epoch in range(num_epochs):
        # Shuffle the data at each epoch
        if shuffle:
            shuffle_indices = np.random.permutation(np.arange(data_size))
            shuffled_data = data[shuffle_indices]
        else:
            shuffled_data = data
        for batch_num in range(num_batches_per_epoch):
            start_index = batch_num * batch_size
            end_index = min((batch_num + 1) * batch_size, data_size)
            yield shuffled_data[start_index:end_index]

x,y=prepare()

#np.random.seed(10)
shuffle_indices = np.random.permutation(np.arange(len(y)))
x_shuffled = x[shuffle_indices]
y_shuffled = y[shuffle_indices]
dev_split=500
x_train, x_dev = x_shuffled[:-dev_split], x_shuffled[-dev_split:]
y_train, y_dev = y_shuffled[:-dev_split], y_shuffled[-dev_split:]
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))

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')

px = tf.placeholder(tf.float32, shape=[None, width*height*channels], name="px")
py = tf.placeholder(tf.float32, shape=[None, classCount], name="py")
keep_prob = tf.placeholder(tf.float32, name="keep_prob")
x_image = tf.reshape(px, [-1,width,height,channels])

W_conv1 = weight_variable([5, 5, channels, 32])
b_conv1 = bias_variable([32])
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([width_c2 * height_c2 * 64, 1024])
b_fc1 = bias_variable([1024])
h_pool2_flat = tf.reshape(h_pool2, [-1, width_c2*height_c2*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, classCount])
b_fc2 = bias_variable([classCount])

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

timestamp = str(int(time.time()))
out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", timestamp))
print("Writing to {}\n".format(out_dir))

global_step = tf.Variable(0, name="global_step", trainable=False)

cross_entropy = tf.reduce_mean(-tf.reduce_sum(py * tf.log(y_conv), reduction_indices=[1]))
train_step = tf.train.AdamOptimizer(1e-5).minimize(cross_entropy, global_step=global_step)
predictions = tf.argmax(y_conv, 1, name="predictions")
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(py,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
sess.run(tf.initialize_all_variables())


checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
    os.makedirs(checkpoint_dir)
saver = tf.train.Saver(tf.all_variables())

batches = batch_iter(list(zip(x_train, y_train)), 100, 20000)
i=0
for batch in batches:
    x_batch, y_batch = zip(*batch)
    x_batch=np.array(x_batch)
    y_batch=np.array(y_batch)
    i+=1

    train_step.run(feed_dict={px: x_batch, py: y_batch, keep_prob: 0.5})

    if i%50 == 0:
        path = saver.save(sess, checkpoint_prefix, global_step=global_step)
        print("\nSaved model checkpoint to {}".format(path))
        print("test accuracy %g\n"%accuracy.eval(feed_dict={px: x_dev, py: y_dev, keep_prob: 1.0}))

    if i%10 == 0:
        train_accuracy = accuracy.eval(feed_dict={px: x_batch, py: y_batch, keep_prob: 1.0})
        print("step %d, training accuracy %g"%(i, train_accuracy))