import tensorflow as tf
import numpy as np
import data.generate_RGB

num_epochs = 5
width = 28
height = 28
num_categories = 3
num_channels = 3
batch_size = 100 # for my sanity
num_training_examples = 5000
num_batches = num_training_examples/batch_size

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')

x = tf.placeholder("float", shape=[None, height, width, num_channels])
x_image = tf.reshape(x, [-1,width,height,num_channels])
y_ = tf.placeholder("float", shape=[None, num_categories])

#1st conv layer
W_conv1 = weight_variable([5, 5, num_channels, 32]) #5x5 conv window, 3 channel, 32 feature maps
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)

#2nd conv layer
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)

#fully connected layer
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)

#droupout
keep_prob = tf.placeholder("float")
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

#softmax output layer
W_fc2 = weight_variable([1024, num_categories])
b_fc2 = bias_variable([num_categories])
y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)

#saving model
saver = tf.train.Saver()

#load dataset
d = data.generate_RGB.GenerateRGB(height, width, 0.05)
dataset = d.generate_data(num_training_examples)
xs = dataset[0]
ys = dataset[1]
xs = np.split(xs, num_batches) #split into 50 minibatches of size (5000/50)=100
ys = np.split(ys, num_batches)

dataset_validation = d.generate_data(500)
xs_v = dataset_validation[0]
ys_v = dataset_validation[1]
xs_v = np.split(xs_v, 50) #split into 50 minibatches of size (500/50)=10 ????
ys_v = np.split(ys_v, 50)

#train and evaluate
with tf.Session() as sess:
    cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))
    train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
    sess.run(tf.initialize_all_variables())
    for j in range(num_epochs):
        for i in range(50):
          train_step.run(feed_dict={x: xs[i], y_: ys[i], keep_prob: 0.5})

        print "=== EPOCH: " + str(j) + " ==="
        print "test accuracy: %g \n"%accuracy.eval(feed_dict={
            x: xs_v[i], y_: ys_v[i], keep_prob: 1.0})

        saver.save(sess, "conv_nets/model.ckpt")