歡迎Follow我的GitHub
本文地址:http://blog.csdn.net/caroline_wendy/article/details/77639394
框架:Python + TensorFlow
知識:工程配置 + HelloWorld + MNIST
本文源碼的GitHub地址
準備
Fork TensorFlow的工程,並下載,轉換遠端Git地址
git remote set-url origin https://github.com/SpikeKing/tensorflow.git
創建Python工程MachineLearningTutorial,使用 virtualenv 創建虛擬環境
pip install virtualenv
virtualenv MLT_ENV
激活或關閉的命令
source MLT_ENV/bin/activate
deactivate
安裝TensorFlow的庫,使用阿里雲的源
pip install TensorFlow -i http://mirrors.aliyun.com/pypi/simple --trusted-host mirrors.aliyun.com
導出庫的版本,及全部安裝
pip freeze>requirements.txt
pip install -r requirements.txt
Hello World
切換Python的解釋器(Interpreter)
HelloWorld
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print sess.run(hello)
a = tf.constant(10)
b = tf.constant(32)
print sess.run(a + b)
輸出
Hello, TensorFlow!
42
避免cpu_feature_guard
警告
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
MNIST
路徑:tensorflow/examples/tutorials/mnist/mnist_softmax.py
加載MNIST數據,默認存放於tmp文件夾,標籤使用one-hot模式
mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)
one-hot的值,如下:[ 0. 0. 0. 0. 0. 0. 0. 1. 0. 0.]
,表示標籤“7”,將類別標籤轉換爲向量,避免標籤的數值關係。
創建變量,placeholder表示輸入數據、Variable表示可變參數,最終公式是y = x * W + b
,y_表示真實標籤(Ground Truth)
x = tf.placeholder(tf.float32, [None, 784])
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
y = tf.matmul(x, W) + b
y_ = tf.placeholder(tf.float32, [None, 10])
計算交叉熵,即損失函數,labels表示真實標籤,logits表示預估標籤
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y))
等價於
sc2 = tf.reduce_sum(-1 * (labels_ * tf.log(tf.nn.softmax(labels))), reduction_indices=[1])
梯度下降的方式,優化交叉熵
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
創建交互會話,初始化全部可變參數
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
每次取出100張圖片,進行訓練,在Session中執行train_step
公式,feed_dict
輸入參數(placeholder),按批次(batch)訓練
for _ in range(1000):
batch_xs, batch_ys = mnist.train.next_batch(100)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
輸出一維列表最大值的索引(argmax),進行比較(equal),再講Bool值轉爲Float(cast),全部求平均(reduce_mean),就是準確率的計算公式
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
在Session中執行accuracy
公式,feed_dict
輸入參數(placeholder),數據源是測試集
print(sess.run(accuracy, feed_dict={x: mnist.test.images,
y_: mnist.test.labels}))
腳本參數是data_dir
,在main函數中,則是FLAGS.data_dir
,默認存放於臨時目錄(tmp),在 tf.app.run()中執行,FLAGS表示指定的參數,如--learning_rate 20
,unparsed表示未指定的參數,隨意輸入的參數。
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--data_dir', type=str, default='/tmp/tensorflow/mnist/input_data',
help='Directory for storing input data')
FLAGS, unparsed = parser.parse_known_args()
tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
Mac系統中,tmp存放隱藏文件,在終端的home目錄中,輸入open /tmp
,即可打開
完整的MNIST代碼,及註釋
FLAGS = None # 全局變量
def main(_):
mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True) # 加載數據源
x = tf.placeholder(tf.float32, [None, 784]) # 數據輸入
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
y = tf.matmul(x, W) + b
y_ = tf.placeholder(tf.float32, [None, 10]) # 標籤輸入
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y)) # 損失函數
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy) # 優化器
sess = tf.InteractiveSession() # 交互會話
tf.global_variables_initializer().run() # 初始化變量
# 訓練模型
for _ in range(1000):
batch_xs, batch_ys = mnist.train.next_batch(100)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
# 驗證模型
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(sess.run(accuracy, feed_dict={x: mnist.test.images,
y_: mnist.test.labels}))
if __name__ == '__main__':
parser = argparse.ArgumentParser() # 設置參數data_dir
parser.add_argument('--data_dir', type=str, default='/tmp/tensorflow/mnist/input_data',
help='Directory for storing input data')
FLAGS, unparsed = parser.parse_known_args()
tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
工程配置 + HelloWorld + MNIST
OK! That’s all! Enjoy it!