4用於cifar10的卷積神經網路-4.6設計模型訓練和評估的會話流程
阿新 • • 發佈:2019-01-07
在TensorFlow中實現這個網路模型
0、載入資料集
1、啟動會話
2、一輪一輪的訓練模型
2.1、在每一輪中分多個批次餵給資料
2.1.1在每個批次上執行訓練節點,訓練模型
2.1.2經過若干個批次後,評估當前的模型,計算訓練集上的損失值,準確率
3、在測試集上評估最終的模型:損失值,準確率
#-*- coding:utf-8 -*-
#實現簡單卷積神經網路對MNIST資料集進行分類:conv2d + activation + pool + fc
import csv
import tensorflow as tf
import os
from tensorflow.examples.tutorials.mnist import input_data
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import sys
from six.moves import urllib
import tarfile
import cifar10_input
import numpy as np
# 設定演算法超引數
learning_rate_init = 0.001
training_epochs = 1
batch_size = 100
display_step = 10
conv1_kernel_num = 32
conv2_kernel_num = 32
# fc1_units_num = 384
fc1_units_num = 32
fc2_units_num = 32
# Network Parameters
n_input = 784 # MNIST data input (img shape: 28*28)
# n_classes = 10 # MNIST total classes (0-9 digits)
#資料集中輸入影象的引數
dataset_dir='../CIFAR10_dataset'
# image_size = 24
# image_channel = 3
# n_classes = 10 #CiFar10中類的數量
num_examples_per_epoch_for_train = cifar10_input.NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN#50000
num_examples_per_epoch_for_eval = cifar10_input.NUM_EXAMPLES_PER_EPOCH_FOR_EVAL#10000
image_size = cifar10_input.IMAGE_SIZE#24
image_channel = 3
n_classes = cifar10_input.NUM_CLASSES #10個分類:CiFar10 中類的數量
#從網址下載資料集存放到data_dir指定的目錄中
def maybe_download_and_extract(data_dir):
"""下載並解壓縮資料集 from Alex's website."""
dest_directory = data_dir
DATA_URL = 'http://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz'
if not os.path.exists(dest_directory):
os.makedirs(dest_directory)
filename = DATA_URL.split('/')[-1] #'cifar-10-binary.tar.gz'
filepath = os.path.join(dest_directory, filename)#'../CIFAR10_dataset\\cifar-10-binary.tar.gz'
if not os.path.exists(filepath):
def _progress(count, block_size, total_size):
sys.stdout.write('\r>> Downloading %s %.1f%%' % (filename,
float(count * block_size) / float(total_size) * 100.0))
sys.stdout.flush()
filepath, _ = urllib.request.urlretrieve(DATA_URL, filepath, _progress)
print()
statinfo = os.stat(filepath)
print('Successfully downloaded', filename, statinfo.st_size, 'bytes.')
extracted_dir_path = os.path.join(dest_directory, 'cifar-10-batches-bin')#'../CIFAR10_dataset\\cifar-10-batches-bin'
if not os.path.exists(extracted_dir_path):
tarfile.open(filepath, 'r:gz').extractall(dest_directory)
def get_distorted_train_batch(data_dir,batch_size):
"""Construct distorted input for CIFAR training using the Reader ops.
Returns:
images: Images. 4D tensor of [batch_size, IMAGE_SIZE, IMAGE_SIZE, 3] size.
labels: Labels. 1D tensor of [batch_size] size.
Raises:
ValueError: If no data_dir
"""
if not data_dir:
raise ValueError('Please supply a data_dir')
data_dir = os.path.join(data_dir, 'cifar-10-batches-bin')
images, labels = cifar10_input.distorted_inputs(data_dir=data_dir,batch_size=batch_size)
return images,labels
def get_undistorted_eval_batch(data_dir,eval_data, batch_size):
"""Construct input for CIFAR evaluation using the Reader ops.
Args:
eval_data: bool, indicating if one should use the train or eval data set.
Returns:
images: Images. 4D tensor of [batch_size, IMAGE_SIZE, IMAGE_SIZE, 3] size.
labels: Labels. 1D tensor of [batch_size] size.
Raises:
ValueError: If no data_dir
"""
if not data_dir:
raise ValueError('Please supply a data_dir')
data_dir = os.path.join(data_dir, 'cifar-10-batches-bin')
images, labels = cifar10_input.inputs(eval_data=eval_data,data_dir=data_dir,batch_size=batch_size)
return images,labels
#根據指定的維數返回初始化好的指定名稱的權重 Variable
def WeightsVariable(shape, name_str, stddev=0.1):
# initial = tf.random_normal(shape=shape, stddev=stddev, dtype=tf.float32)
initial = tf.truncated_normal(shape=shape, stddev=stddev, dtype=tf.float32)
return tf.Variable(initial, dtype=tf.float32, name=name_str)
#根據指定的維數返回初始化好的指定名稱的偏置 Variable
def BiasesVariable(shape, name_str, init_value=0.00001):
initial = tf.constant(init_value, shape=shape)
return tf.Variable(initial, dtype=tf.float32, name=name_str)
# 二維卷積層activation(conv2d+bias)的封裝
def Conv2d(x, W, b, stride=1, padding='SAME',activation=tf.nn.relu,act_name='relu'):
with tf.name_scope('conv2d_bias'):
y = tf.nn.conv2d(x, W, strides=[1, stride, stride, 1], padding=padding)
y = tf.nn.bias_add(y, b)
with tf.name_scope(act_name):
y = activation(y)
return y
# 二維池化層pool的封裝
def Pool2d(x, pool= tf.nn.max_pool, k=2, stride=2,padding='SAME'):
return pool(x, ksize=[1, k, k, 1], strides=[1, stride, stride, 1], padding=padding)
# 全連線層activate(wx+b)的封裝
def FullyConnected(x, W, b, activate=tf.nn.relu, act_name='relu'):
with tf.name_scope('Wx_b'):
y = tf.matmul(x, W)
y = tf.add(y, b)
with tf.name_scope(act_name):
y = activate(y)
return y
def Inference(image_holder):
# 第一個卷積層activate(conv2d + biase)
with tf.name_scope('Conv2d_1'):
# conv1_kernel_num = 64
weights = WeightsVariable(shape=[5, 5, image_channel, conv1_kernel_num],
name_str='weights',stddev=5e-2)
biases = BiasesVariable(shape=[conv1_kernel_num], name_str='biases',init_value=0.0)
conv1_out = Conv2d(image_holder, weights, biases, stride=1, padding='SAME')
# 第一個池化層(pool 2d)
with tf.name_scope('Pool2d_1'):
pool1_out = Pool2d(conv1_out, pool=tf.nn.max_pool, k=3, stride=2,padding='SAME')
# 第二個卷積層activate(conv2d + biase)
with tf.name_scope('Conv2d_2'):
# conv2_kernels_num = 64
weights = WeightsVariable(shape=[5, 5, conv1_kernel_num, conv2_kernel_num],
name_str='weights', stddev=5e-2)
biases = BiasesVariable(shape=[conv2_kernel_num], name_str='biases', init_value=0.0)
conv2_out = Conv2d(pool1_out, weights, biases, stride=1, padding='SAME')
# 第二個池化層(pool 2d)
with tf.name_scope('Pool2d_2'):
pool2_out = Pool2d(conv2_out, pool=tf.nn.max_pool, k=3, stride=2, padding='SAME')
#將二維特徵圖變換為一維特徵向量
with tf.name_scope('FeatsReshape'):
features = tf.reshape(pool2_out, [batch_size,-1])
feats_dim = features.get_shape()[1].value
# 第一個全連線層(fully connected layer)
with tf.name_scope('FC1_nonlinear'):
# fc1_units_num = 384
weights = WeightsVariable(shape=[feats_dim, fc1_units_num],
name_str='weights',stddev=4e-2)
biases = BiasesVariable(shape=[fc1_units_num], name_str='biases',init_value=0.1)
fc1_out = FullyConnected(features, weights, biases, activate=tf.nn.relu, act_name='relu')
# 第二個全連線層(fully connected layer)
with tf.name_scope('FC2_nonlinear'):
# fc2_units_num = 192
weights = WeightsVariable(shape=[fc1_units_num, fc2_units_num],
name_str='weights',stddev=4e-2)
biases = BiasesVariable(shape=[fc2_units_num], name_str='biases',init_value=0.1)
fc2_out = FullyConnected(fc1_out, weights, biases,activate=tf.nn.relu, act_name='relu')
# 第三個全連線層(fully connected layer)
with tf.name_scope('FC3_linear'):
fc3_units_num = n_classes
weights = WeightsVariable(shape=[fc2_units_num, fc3_units_num],
name_str='weights',stddev=1.0/fc2_units_num)
biases = BiasesVariable(shape=[fc3_units_num], name_str='biases',init_value=0.0)
logits = FullyConnected(fc2_out, weights, biases,activate=tf.identity, act_name='linear')
return logits
def TrainModel():
#呼叫上面寫的函式構造計算圖
with tf.Graph().as_default():
# 計算圖輸入
with tf.name_scope('Inputs'):
image_holder = tf.placeholder(tf.float32, [batch_size, image_size,image_size,image_channel], name='images')
labels_holder = tf.placeholder(tf.int32, [batch_size], name='labels')
# 計算圖前向推斷過程
with tf.name_scope('Inference'):
logits = Inference(image_holder)
# 定義損失層(loss layer)
with tf.name_scope('Loss'):
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=labels_holder,logits=logits)
cross_entropy_mean = tf.reduce_mean(cross_entropy)
total_loss = cross_entropy_mean
# 定義優化訓練層(train layer)
with tf.name_scope('Train'):
learning_rate = tf.placeholder(tf.float32)
global_step = tf.Variable(0, name='global_step', trainable=False, dtype=tf.int64)
optimizer = tf.train.RMSPropOptimizer(learning_rate=learning_rate)
train_op = optimizer.minimize(total_loss,global_step=global_step)
# 定義模型評估層(evaluate layer)
with tf.name_scope('Evaluate'):
top_K_op = tf.nn.in_top_k(predictions=logits,targets=labels_holder,k=1)
#定義獲取訓練樣本批次的計算節點
with tf.name_scope('GetTrainBatch'):
image_train,labels_train = get_distorted_train_batch(data_dir=dataset_dir,batch_size=batch_size)
# 定義獲取測試樣本批次的計算節點
with tf.name_scope('GetTestBatch'):
image_test, labels_test = get_undistorted_eval_batch(eval_data=True,data_dir=dataset_dir, batch_size=batch_size)
# 新增所有變數的初始化節點
init_op = tf.global_variables_initializer()
print('把計算圖寫入事件檔案,在TensorBoard裡面檢視')
graph_writer = tf.summary.FileWriter(logdir='logs', graph=tf.get_default_graph())
graph_writer.close()
# 將評估結果儲存到檔案
results_list = list()
# 寫入引數配置
results_list.append(['learning_rate', learning_rate_init,
'training_epochs', training_epochs,
'batch_size', batch_size,
'conv1_kernel_num', conv1_kernel_num,
'conv2_kernel_num', conv2_kernel_num,
'fc1_units_num', fc1_units_num,
'fc2_units_num', fc2_units_num])
results_list.append(['train_step', 'train_loss','train_step', 'train_accuracy'])
with tf.Session() as sess:
sess.run(init_op)
print('===>>>>>>>==開始訓練集上訓練模型==<<<<<<<=====')
total_batches = int(num_examples_per_epoch_for_train / batch_size)
print('Per batch Size:,',batch_size)
print('Train sample Count Per Epoch:',num_examples_per_epoch_for_train)
print('Total batch Count Per Epoch:', total_batches)
#啟動資料讀取佇列
tf.train.start_queue_runners()
#記錄模型被訓練的步數
training_step = 0
# 訓練指定輪數,每一輪的訓練樣本總數為:num_examples_per_epoch_for_train
for epoch in range(training_epochs):
#每一輪都要把所有的batch跑一遍
for batch_idx in range(total_batches):
#執行獲取訓練資料的計算圖,取出一個批次資料
images_batch ,labels_batch = sess.run([image_train,labels_train])
#執行優化器訓練節點
_,loss_value = sess.run([train_op,total_loss],
feed_dict={image_holder:images_batch,
labels_holder:labels_batch,
learning_rate:learning_rate_init})
#每呼叫一次訓練節點,training_step就加1,最終==training_epochs * total_batch
training_step = sess.run(global_step)
#每訓練display_step次,計算當前模型的損失和分類準確率
if training_step % display_step == 0:
#執行accuracy節點,計算當前批次的訓練樣本的準確率
predictions = sess.run([top_K_op],
feed_dict={image_holder:images_batch,
labels_holder:labels_batch})
#當前批次上的預測正確的樣本量
batch_accuracy = np.sum(predictions)/batch_size
results_list.append([training_step,loss_value,training_step,batch_accuracy])
print("Training Step:" + str(training_step) +
",Training Loss = " + "{:.6f}".format(loss_value) +
",Training Accuracy = " + "{:.5f}".format(batch_accuracy) )
print('訓練完畢')
print('===>>>>>>>==開始在測試集上評估模型==<<<<<<<=====')
total_batches = int(num_examples_per_epoch_for_eval / batch_size)
total_examples = total_batches * batch_size
print('Per batch Size:,', batch_size)
print('Test sample Count Per Epoch:', total_examples)
print('Total batch Count Per Epoch:', total_batches)
correct_predicted = 0
for test_step in range(total_batches):
#執行獲取測試資料的計算圖,取出一個批次測試資料
images_batch,labels_batch = sess.run([image_test,labels_test])
#執行accuracy節點,計算當前批次的測試樣本的準確率
predictions = sess.run([top_K_op],
feed_dict={image_holder:images_batch,
labels_holder:labels_batch})
#累計每個批次上的預測正確的樣本量
correct_predicted += np.sum(predictions)
accuracy_score = correct_predicted / total_examples
print('---------->Accuracy on Test Examples:',accuracy_score)
results_list.append(['Accuracy on Test Examples:',accuracy_score])
# 將評估結果儲存到檔案
results_file = open('evaluate_results.csv', 'w', newline='')
csv_writer = csv.writer(results_file, dialect='excel')
for row in results_list:
csv_writer.writerow(row)
def main(argv=None):
maybe_download_and_extract(data_dir=dataset_dir)
train_dir='train/'
if tf.gfile.Exists(train_dir):
tf.gfile.DeleteRecursively(train_dir)
tf.gfile.MakeDirs(train_dir)
TrainModel()
if __name__ =='__main__':
tf.app.run()
輸出:
把計算圖寫入事件檔案,在TensorBoard裡面檢視
===>>>>>>>==開始訓練集上訓練模型==<<<<<<<=====
Per batch Size:, 100
Train sample Count Per Epoch: 50000
Total batch Count Per Epoch: 500
Training Step:10,Training Loss = 2.306586,Training Accuracy = 0.06000
Training Step:20,Training Loss = 2.306300,Training Accuracy = 0.06000
Training Step:30,Training Loss = 2.303389,Training Accuracy = 0.10000
Training Step:40,Training Loss = 2.302423,Training Accuracy = 0.09000
Training Step:50,Training Loss = 2.302983,Training Accuracy = 0.14000
Training Step:60,Training Loss = 2.301472,Training Accuracy = 0.14000
Training Step:70,Training Loss = 2.306488,Training Accuracy = 0.08000
Training Step:80,Training Loss = 2.303758,Training Accuracy = 0.17000
Training Step:90,Training Loss = 2.301871,Training Accuracy = 0.19000
Training Step:100,Training Loss = 2.299527,Training Accuracy = 0.12000
Training Step:110,Training Loss = 2.298491,Training Accuracy = 0.15000
Training Step:120,Training Loss = 2.286592,Training Accuracy = 0.20000
Training Step:130,Training Loss = 2.202128,Training Accuracy = 0.22000
Training Step:140,Training Loss = 2.026917,Training Accuracy = 0.25000
Training Step:150,Training Loss = 2.103927,Training Accuracy = 0.22000
Training Step:160,Training Loss = 2.094987,Training Accuracy = 0.25000
Training Step:170,Training Loss = 1.953211,Training Accuracy = 0.32000
Training Step:180,Training Loss = 1.974621,Training Accuracy = 0.27000
Training Step:190,Training Loss = 1.944889,Training Accuracy = 0.29000
Training Step:200,Training Loss = 1.896667,Training Accuracy = 0.32000
Training Step:210,Training Loss = 2.002789,Training Accuracy = 0.25000
Training Step:220,Training Loss = 2.057821,Training Accuracy = 0.29000
Training Step:230,Training Loss = 1.891200,Training Accuracy = 0.33000
Training Step:240,Training Loss = 1.877370,Training Accuracy = 0.33000
Training Step:250,Training Loss = 1.904065,Training Accuracy = 0.34000
Training Step:260,Training Loss = 1.795470,Training Accuracy = 0.41000
Training Step:270,Training Loss = 1.956472,Training Accuracy = 0.35000
Training Step:280,Training Loss = 1.893711,Training Accuracy = 0.30000
Training Step:290,Training Loss = 1.737094,Training Accuracy = 0.34000
Training Step:300,Training Loss = 1.759201,Training Accuracy = 0.43000
Training Step:310,Training Loss = 2.055783,Training Accuracy = 0.33000
Training Step:320,Training Loss = 1.666109,Training Accuracy = 0.42000
Training Step:330,Training Loss = 1.816121,Training Accuracy = 0.32000
Training Step:340,Training Loss = 1.806642,Training Accuracy = 0.41000
Training Step:350,Training Loss = 1.779170,Training Accuracy = 0.35000
Training Step:360,Training Loss = 1.755931,Training Accuracy = 0.45000
Training Step:370,Training Loss = 1.692869,Training Accuracy = 0.43000
Training Step:380,Training Loss = 1.975068,Training Accuracy = 0.34000
Training Step:390,Training Loss = 1.735186,Training Accuracy = 0.42000
Training Step:400,Training Loss = 1.651298,Training Accuracy = 0.43000
Training Step:410,Training Loss = 1.725129,Training Accuracy = 0.45000
Training Step:420,Training Loss = 1.673964,Training Accuracy = 0.43000
Training Step:430,Training Loss = 1.848086,Training Accuracy = 0.37000
Training Step:440,Training Loss = 1.689408,Training Accuracy = 0.41000
Training Step:450,Training Loss = 1.647934,Training Accuracy = 0.41000
Training Step:460,Training Loss = 1.587230,Training Accuracy = 0.43000
Training Step:470,Training Loss = 1.714040,Training Accuracy = 0.47000
Training Step:480,Training Loss = 1.771270,Training Accuracy = 0.49000
Training Step:490,Training Loss = 1.757897,Training Accuracy = 0.38000
Training Step:500,Training Loss = 1.656613,Training Accuracy = 0.41000
訓練完畢
===>>>>>>>==開始在測試集上評估模型==<<<<<<<=====
Per batch Size:, 100
Test sample Count Per Epoch: 10000
Total batch Count Per Epoch: 100
---------->Accuracy on Test Examples: 0.4365