引言
TensorFlow 是 Google 基于 DistBelief 进行研发的第二代人工智能学习系统,被广泛用于语音识别或图像识别等多项机器深度学习领域。其命名来源于本身的运行原理。Tensor(张量)意味着 N 维数组,Flow(流)意味着基于数据流图的计算,TensorFlow 代表着张量从图象的一端流动到另一端计算过程,是将复杂的数据结构传输至人工智能神经网中进行分析和处理的过程。
TensorFlow 完全开源,任何人都可以使用。可在小到一部智能手机、大到数千台数据中心服务器的各种设备上运行。
『机器学习进阶笔记』系列是将深入解析 TensorFlow 系统的技术实践,从零开始,由浅入深,与大家一起走上机器学习的进阶之路。
CUDA 与 TensorFlow 安装
按以往经验,TensorFlow 安装一条 pip 命令就可以解决,前提是有 fq 工具,没有的话去找找墙内别人分享的地址。而坑多在安装支持 gpu,需预先安装英伟达的 cuda,这里坑比较多,推荐使用 ubuntu deb 的安装方式来安装 cuda,run.sh 的方式总感觉有很多问题,cuda 的安装具体可以参考。 注意链接里面的 tensorflow 版本是以前的,tensorflow 现在官方上的要求是 cuda7.5+cudnnV4,请在安装的时候注意下。
Hello World
import tensorflow as tfhello = tf.constant('Hello, TensorFlow!')sess = tf.Session()print sess.run(hello)
首先,通过 tf.constant 创建一个常量,然后启动 Tensorflow 的 Session,调用 sess 的 run 方法来启动整个 graph。
接下来我们做下简单的数学的方法:
import tensorflow as tfa = tf.constant(2)b = tf.constant(3)with tf.Session() as sess:print "a=2, b=3"print "Addition with constants: %i" % sess.run(a+b)print "Multiplication with constants: %i" % sess.run(a*b)# outputa=2, b=3Addition with constants: 5Multiplication with constants: 6
接下来用 tensorflow 的 placeholder 来定义变量做类似计算:
placeholder 的使用见https://www.tensorflow.org/versions/r0.8/api_docs/python/io_ops.html#placeholder
import tensorflow as tfa = tf.placeholder(tf.int16)b = tf.placeholder(tf.int16)add = tf.add(a, b)mul = tf.mul(a, b)with tf.Session() as sess:# Run every operation with variable inputprint "Addition with variables: %i" % sess.run(add, feed_dict={a: 2, b: 3})print "Multiplication with variables: %i" % sess.run(mul, feed_dict={a: 2, b: 3})# output:Addition with variables: 5Multiplication with variables: 6matrix1 = tf.constant([[3., 3.]])matrix2 = tf.constant([[2.],[2.]])with tf.Session() as sess:result = sess.run(product)print result
线性回归
以下代码来自GitHub - aymericdamien/TensorFlow-Examples: TensorFlow Tutorial and Examples for beginners, 仅作学习用
import tensorflow as tfimport numpyimport matplotlib.pyplot as pltrng = numpy.random# Parameterslearning_rate = 0.01training_epochs = 2000display_step = 50# Training Datatrain_X = numpy.asarray([3.3,4.4,5.5,6.71,6.93,4.168,9.779,6.182,7.59,2.167,7.042,10.791,5.313,7.997,5.654,9.27,3.1])train_Y = numpy.asarray([1.7,2.76,2.09,3.19,1.694,1.573,3.366,2.596,2.53,1.221,2.827,3.465,1.65,2.904,2.42,2.94,1.3])n_samples = train_X.shape[0]# tf Graph InputX = tf.placeholder("float")Y = tf.placeholder("float")# Create Model# Set model weightsW = tf.Variable(rng.randn(), name="weight")b = tf.Variable(rng.randn(), name="bias")# Construct a linear modelactivation = tf.add(tf.mul(X, W), b)# Minimize the squared errorscost = tf.reduce_sum(tf.pow(activation-Y, 2))/(2*n_samples) #L2 lossoptimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost) #Gradient descent# Initializing the variablesinit = tf.initialize_all_variables()# Launch the graphwith tf.Session() as sess:sess.run(init)# Fit all training datafor epoch in range(training_epochs):for (x, y) in zip(train_X, train_Y):sess.run(optimizer, feed_dict={X: x, Y: y})#Display logs per epoch stepif epoch % display_step == 0:print "Epoch:", '%04d' % (epoch+1), "cost=", \"{:.9f}".format(sess.run(cost, feed_dict={X: train_X, Y:train_Y})), \"W=", sess.run(W), "b=", sess.run(b)print "Optimization Finished!"print "cost=", sess.run(cost, feed_dict={X: train_X, Y: train_Y}), \"W=", sess.run(W), "b=", sess.run(b)#Graphic displayplt.plot(train_X, train_Y, 'ro', label='Original data')plt.plot(train_X, sess.run(W) * train_X + sess.run(b), label='Fitted line')plt.legend()plt.show()
逻辑回归
import tensorflow as tf# Import MINST datafrom tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets("/tmp/data/", one_hot=True)# Parameterslearning_rate = 0.01training_epochs = 25batch_size = 100display_step = 1# tf Graph Inputx = tf.placeholder(tf.float32, [None, 784]) # mnist data image of shape 28*28=784y = tf.placeholder(tf.float32, [None, 10]) # 0-9 digits recognition => 10 classes# Set model weightsW = tf.Variable(tf.zeros([784, 10]))b = tf.Variable(tf.zeros([10]))# Construct modelpred = tf.nn.softmax(tf.matmul(x, W) + b) # Softmax# Minimize error using cross entropycost = tf.reduce_mean(-tf.reduce_sum(y*tf.log(pred), reduction_indices=1))# Gradient Descentoptimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)# Initializing the variablesinit = tf.initialize_all_variables()# Launch the graphwith tf.Session() as sess:sess.run(init)# Training cyclefor epoch in range(training_epochs):avg_cost = 0.total_batch = int(mnist.train.num_examples/batch_size)# Loop over all batchesfor i in range(total_batch):batch_xs, batch_ys = mnist.train.next_batch(batch_size)# Run optimization op (backprop) and cost op (to get loss value)_, c = sess.run([optimizer, cost], feed_dict={x: batch_xs,y: batch_ys})# Compute average lossavg_cost += c / total_batch# Display logs per epoch stepif (epoch+1) % display_step == 0:print "Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(avg_cost)print "Optimization Finished!"# Test modelcorrect_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))# Calculate accuracyaccuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))print "Accuracy:", accuracy.eval({x: mnist.test.images, y: mnist.test.labels})# result :Epoch: 0001 cost= 29.860467369Epoch: 0002 cost= 22.001451784Epoch: 0003 cost= 21.019925554Epoch: 0004 cost= 20.561320320Epoch: 0005 cost= 20.109135756Epoch: 0006 cost= 19.927862290Epoch: 0007 cost= 19.548687116Epoch: 0008 cost= 19.429119071Epoch: 0009 cost= 19.397068211Epoch: 0010 cost= 19.180813479Epoch: 0011 cost= 19.026808132Epoch: 0012 cost= 19.057875510Epoch: 0013 cost= 19.009575057Epoch: 0014 cost= 18.873240641Epoch: 0015 cost= 18.718575359Epoch: 0016 cost= 18.718761925Epoch: 0017 cost= 18.673640560Epoch: 0018 cost= 18.562128253Epoch: 0019 cost= 18.458205289Epoch: 0020 cost= 18.538211225Epoch: 0021 cost= 18.443384213Epoch: 0022 cost= 18.428727668Epoch: 0023 cost= 18.304270616Epoch: 0024 cost= 18.323529782Epoch: 0025 cost= 18.247192113Optimization Finished!(10000, 784)Accuracy 0.9206
这里有个小插曲,ipython notebook 在一个 notebook 打开时,一直在占用 GPU 资源,可能是之前有一个 notebook 一直打开着,然后占用着 GPU 资源,然后在计算 Accuracy 的”InternalError: Dst tensor is not initialized.” 然后找了 github 上面也有这个问题InternalError: Dst tensor is not initialized., 可以肯定是 GPU 的 memory 相关的问题,所以就尝试加上 tf.device(‘/cpu:0’),将 Accuracy 这步拉到 cpu 上计算,但是又出现 OOM 的问题,最后 nvidia-smi 时,发现有一个 python 脚本一直占用 3g 多的显存,把它 kill 之后恢复了,之前还比较吐槽怎么可能 10000*784 个 float 就把显存撑爆呢,原来是自己的问题。
这里逻辑回归,model 是一个 softmax 函数用来做多元分类,大概意思是选择 10 当中最后预测概率最高作为最终的分类。
其实基本的 tensorflow 没有特别好讲的,语法的课程什么可以去看看基本的文档,之后我会找一点经典有趣的 tensorflow 的代码应用来看看,毕竟『show me the code 』才是程序猿应有的态度。
本文由『UCloud 内核与虚拟化研发团队』提供。
关于作者:
Burness( ), UCloud 平台研发中心深度学习研发工程师,tflearn Contributor,做过电商推荐、精准化营销相关算法工作,专注于分布式深度学习框架、计算机视觉算法研究,平时喜欢玩玩算法,研究研究开源的项目,偶尔也会去一些数据比赛打打酱油,生活中是个极客,对新技术、新技能痴迷。
「UCloud 机构号」将独家分享云计算领域的技术洞见、行业资讯以及一切你想知道的相关讯息。
欢迎提问 & 求关注 o(////▽////)q~
