siamese

gistfile1.txt

""" Siamese implementation using Tensorflow with MNIST example.
This siamese network embeds a 28x28 image (a point in 784D) 
into a point in 2D.

By Youngwook Paul Kwon (young at berkeley.edu)
"""

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

#import system things
from tensorflow.examples.tutorials.mnist import input_data # for data
import tensorflow as tf
import numpy as np
import os

#import helpers
import inference
import visualize

# prepare data and tf.session
# mnist = input_data.read_data_sets('MNIST_data', one_hot=False)
sketch_test = np.load("test_set.npy")
sketch_test_label = np.load("test_set_label.npy")
sketch_training = np.load("train_set.npy")
sketch_training_label = np.load("train_set_label.npy")

sess = tf.InteractiveSession()

# setup siamese network
siamese = inference.siamese();
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(siamese.loss)
saver = tf.train.Saver()
tf.initialize_all_variables().run()

# if you just want to load a previously trainmodel?
new = True
model_ckpt = 'model.ckpt'
if os.path.isfile(model_ckpt):
    input_var = None
    while input_var not in ['yes', 'no']:
        input_var = raw_input("We found model.ckpt file. Do you want to load it [yes/no]?")
    if input_var == 'yes':
        new = False

def getBatch(length, num, i):
    return {'start':(length/num)*(i%num), 'end':(length/num)*((i+1)%num)-1, 'no': i%num, 'epoch':int(i/num)}
        
        
# start training
if new:
    i=0
    for step in range(1100):
        #batch_x1, batch_y1 = mnist.train.next_batch(128)
        #batch_x2, batch_y2 = mnist.train.next_batch(128)
        #batch_y = (batch_y1 == batch_y2).astype('float')

        batch1 = getBatch(length, 5, i);
        batch2 = getBatch(length, 5, i);
        i++;
    
        batch_x1 = trainX[batch1['start']:batch1['end'], ]
        batch_y1 = trainY[batch1['start']:batch1['end'], ]
        
        batch_x2 = trainX[batch2['start']:batch2['end'], ]
        batch_y2 = trainY[batch2['start']:batch2['end'], ]
        
        batch_y = (batch_y1 == batch_y2).astype('float');
        
        _, loss_v = sess.run([train_step, siamese.loss], feed_dict={
                            siamese.x1: batch_x1, 
                            siamese.x2: batch_x2, 
                            siamese.y_: batch_y})

        if np.isnan(loss_v):
            print('Model diverged with loss = NaN')
            quit()

        if step % 10 == 0:
            print ('step %d: loss %.3f' % (step, loss_v))

        if step % 1000 == 0 and step > 0:
            saver.save(sess, 'model.ckpt')
            embed = siamese.o1.eval({siamese.x1: mnist.test.images})
            embed.tofile('embed.txt')
else:
    saver.restore(sess, 'model.ckpt')

# visualize result
x_test = mnist.test.images.reshape([-1, 28, 28])
visualize.visualize(embed, x_test)