In [1]:
import os
import random
import numpy as np
import tensorflow as tf
import datetime
from PIL import Image
# 用前10000个数据进行10-fold的cross_validation
dir = "/media/gui/LENOVO/学习/AMAZON/train-jpg/"
label_name = "cloudy"
kp = 0.7 # dropout = 1-kp
def read_label(filname,label_name):
with open("/media/gui/LENOVO/学习/AMAZON/train_v2.csv/train_v2.csv") as fl:
raw = fl.readlines()
fl.close()
result = list()
for part in raw:
line = part.strip("\n").split(",")
mutilabels = line[1].strip(" ")
if label_name in mutilabels:
result.append([1,0])
else:
result.append([0,1])
return result
all_input_label = read_label("/media/gui/LENOVO/学习/AMAZON/train_v2.csv/train_v2.csv",label_name)
# 定义输入节点，对应于图片像素值矩阵集合和图片标签(即所代表的数字)
x = tf.placeholder(tf.float32, shape=[None, 64*64*3])
y_ = tf.placeholder(tf.float32, shape=[None, 2])
x_image = tf.reshape(x, [-1, 64, 64, 3])
# 定义第一个卷积层的variables和ops
W_conv1 = tf.Variable(tf.truncated_normal([5, 5, 3, 16], stddev=0.1))
b_conv1 = tf.Variable(tf.constant(0.1, shape=[16]))
L1_conv = tf.nn.conv2d(x_image, W_conv1, strides=[1, 1, 1, 1], padding='SAME')
L1_relu = tf.nn.relu(L1_conv + b_conv1)
L1_pool = tf.nn.max_pool(L1_relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
# 定义第二个卷积层的variables和ops
W_conv2 = tf.Variable(tf.truncated_normal([3, 3, 16, 32], stddev=0.1))
b_conv2 = tf.Variable(tf.constant(0.1, shape=[32]))
L2_conv = tf.nn.conv2d(L1_pool, W_conv2, strides=[1, 1, 1, 1], padding='SAME')
L2_relu = tf.nn.relu(L2_conv + b_conv2)
L2_pool = tf.nn.max_pool(L2_relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
# 全连接层
W_fc1 = tf.Variable(tf.truncated_normal([16 * 16 * 32, 256], stddev=0.1))
b_fc1 = tf.Variable(tf.constant(0.1, shape=[256]))
h_pool2_flat = tf.reshape(L2_pool, [-1, 16*16*32])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
# dropout
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
# readout层
W_fc2 = tf.Variable(tf.truncated_normal([256, 2], stddev=0.1))
b_fc2 = tf.Variable(tf.constant(0.1, shape=[2]))
y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
# 定义优化器和训练op
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
# 加入正则化
tf.add_to_collection(tf.GraphKeys.WEIGHTS, W_conv1)
tf.add_to_collection(tf.GraphKeys.WEIGHTS, W_conv2)
tf.add_to_collection(tf.GraphKeys.WEIGHTS, W_fc1)
tf.add_to_collection(tf.GraphKeys.WEIGHTS, W_fc2)
regularizer = tf.contrib.layers.l2_regularizer(scale=5.0/50000)
reg_term = tf.contrib.layers.apply_regularization(regularizer)
cross_entropy = (cross_entropy + reg_term)
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, tf.float32))
file_prefix = "/media/gui/LENOVO/学习/AMAZON/project/train-jpg-all-"
data_iteration_num = 3
split_count = 1000
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
print("Start the tensorflow session")
'''设置每次训练op的输入个数和迭代次数，这里为了支持任意图片总数，定义了一个余数remainder，
譬如，如果每次训练op的输入个数为60，图片总数为150张，则前面两次各输入60张，最后一次输入30张（余数30）'''
batch_size = 50
iterations = 100
for bi in range(data_iteration_num):
train_data_index = random.randint(0, 8)
input_images = np.loadtxt(file_prefix + str(train_data_index), dtype="int")
# 执行训练迭代
for it in range(iterations):
t_pre = datetime.datetime.now()
# print("Start the "+str(it+1)+" iteration.")
# for sp in range(1):
# input_images = np.loadtxt( file_prefix+str(sp),dtype = "int")
input_count = len(input_images)
batches_count = int(input_count / batch_size)
remainder = input_count % batch_size
input_labels = all_input_label[(train_data_index*split_count+0):(train_data_index*split_count+input_count)]
for n in range(batches_count):
train_step.run(feed_dict={x: input_images[n * batch_size:(n + 1) * batch_size],
y_: input_labels[n * batch_size:(n + 1) * batch_size], keep_prob: kp})
# print("Running the " + str(n + 1) + " batch in " + str(it + 1) + " iteration.")
if remainder > 0:
start_index = batches_count * batch_size
train_step.run(feed_dict={x: input_images[start_index:input_count - 1],
y_: input_labels[start_index:input_count - 1], keep_prob: kp})
iterate_accuracy = 0
if (it + 1) % 20 == 0:
iterate_accuracy = accuracy.eval(feed_dict={x: input_images, y_: input_labels, keep_prob: 1.0})
print('iteration %d-%d: accuracy %s' % (bi,it + 1, iterate_accuracy))
if iterate_accuracy >= 1:
break
test_images = np.loadtxt(file_prefix + str(9))
test_labels = all_input_label[9000:10000]
print("The end of %d out-iteration, The test accuracy is as follows." % bi)
print("test accuracy %g" % accuracy.eval(feed_dict={x: test_images, y_: test_labels, keep_prob: 1.0}))
# t_after = datetime.datetime.now()
# t = (t_after-t_pre).seconds
# print("Have costed "+str(t)+" in last iteration.")
test_images = np.loadtxt(file_prefix + str(9))
test_labels = input_labels = all_input_label[9000:10000]
print("Final test accuracy %g" % accuracy.eval(feed_dict={x: test_images, y_: test_labels, keep_prob: 1.0}))
print("end")