我們需要評(píng)估模型預(yù)測(cè)值來評(píng)估訓(xùn)練的好壞。

模型評(píng)估是非常重要的，隨后的每個(gè)模型都有模型評(píng)估方式。使用TensorFlow時(shí)，需要把模型評(píng)估加入到計(jì)算圖中，然后在模型訓(xùn)練完后調(diào)用模型評(píng)估。

在訓(xùn)練模型過程中，模型評(píng)估能洞察模型算法，給出提示信息來調(diào)試、提高或者改變整個(gè)模型。但是在模型訓(xùn)練中并不是總需要模型評(píng)估，我們將展示如何在回歸算法和分類算法中使用它。

訓(xùn)練模型之后，需要定量評(píng)估模型的性能如何。在理想情況下，評(píng)估模型需要一個(gè)訓(xùn)練數(shù)據(jù)集和測(cè)試數(shù)據(jù)集，有時(shí)甚至需要一個(gè)驗(yàn)證數(shù)據(jù)集。

想評(píng)估一個(gè)模型時(shí)就得使用大批量數(shù)據(jù)點(diǎn)。如果完成批量訓(xùn)練，我們可以重用模型來預(yù)測(cè)批量數(shù)據(jù)點(diǎn)。但是如果要完成隨機(jī)訓(xùn)練，就不得不創(chuàng)建單獨(dú)的評(píng)估器來處理批量數(shù)據(jù)點(diǎn)。

分類算法模型基于數(shù)值型輸入預(yù)測(cè)分類值，實(shí)際目標(biāo)是1和0的序列。我們需要度量預(yù)測(cè)值與真實(shí)值之間的距離。分類算法模型的損失函數(shù)一般不容易解釋模型好壞，所以通常情況是看下準(zhǔn)確預(yù)測(cè)分類的結(jié)果的百分比。

不管算法模型預(yù)測(cè)的如何，我們都需要測(cè)試算法模型，這點(diǎn)相當(dāng)重要。在訓(xùn)練數(shù)據(jù)和測(cè)試數(shù)據(jù)上都進(jìn)行模型評(píng)估，以搞清楚模型是否過擬合。

# TensorFlowm模型評(píng)估## This code will implement two models. The first# is a simple regression model, we will show how to# call the loss function, MSE during training, and# output it after for test and training sets.## The second model will be a simple classification# model. We will also show how to print percent# classified correctly during training and after# for both the test and training sets.import matplotlib.pyplot as pltimport numpy as npimport tensorflow as tffrom tensorflow.python.framework import opsops.reset_default_graph()# 創(chuàng)建計(jì)算圖sess = tf.Session()# 回歸例子:# We will create sample data as follows:# x-data: 100 random samples from a normal ~ N(1, 0.1)# target: 100 values of the value 10.# We will fit the model:# x-data * A = target# 理論上, A = 10.# 聲明批量大小batch_size = 25# 創(chuàng)建數(shù)據(jù)集x_vals = np.random.normal(1, 0.1, 100)y_vals = np.repeat(10., 100)x_data = tf.placeholder(shape=[None, 1], dtype=tf.float32)y_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)# 八二分訓(xùn)練/測(cè)試數(shù)據(jù) train/test = 80%/20%train_indices = np.random.choice(len(x_vals), round(len(x_vals)*0.8), replace=False)test_indices = np.array(list(set(range(len(x_vals))) - set(train_indices)))x_vals_train = x_vals[train_indices]x_vals_test = x_vals[test_indices]y_vals_train = y_vals[train_indices]y_vals_test = y_vals[test_indices]# 創(chuàng)建變量 (one model parameter = A)A = tf.Variable(tf.random_normal(shape=[1,1]))# 增加操作到計(jì)算圖my_output = tf.matmul(x_data, A)# 增加L2損失函數(shù)到計(jì)算圖loss = tf.reduce_mean(tf.square(my_output - y_target))# 創(chuàng)建優(yōu)化器my_opt = tf.train.GradientDescentOptimizer(0.02)train_step = my_opt.minimize(loss)# 初始化變量init = tf.global_variables_initializer()sess.run(init)# 迭代運(yùn)行# 如果在損失函數(shù)中使用的模型輸出結(jié)果經(jīng)過轉(zhuǎn)換操作，例如，sigmoid_cross_entropy_with_logits（）函數(shù)，# 為了精確計(jì)算預(yù)測(cè)結(jié)果，別忘了在模型評(píng)估中也要進(jìn)行轉(zhuǎn)換操作。for i in range(100):  rand_index = np.random.choice(len(x_vals_train), size=batch_size)  rand_x = np.transpose([x_vals_train[rand_index]])  rand_y = np.transpose([y_vals_train[rand_index]])  sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})  if (i+1)%25==0:    print('Step #' + str(i+1) + ' A = ' + str(sess.run(A)))    print('Loss = ' + str(sess.run(loss, feed_dict={x_data: rand_x, y_target: rand_y})))# 評(píng)估準(zhǔn)確率(loss)mse_test = sess.run(loss, feed_dict={x_data: np.transpose([x_vals_test]), y_target: np.transpose([y_vals_test])})mse_train = sess.run(loss, feed_dict={x_data: np.transpose([x_vals_train]), y_target: np.transpose([y_vals_train])})print('MSE on test:' + str(np.round(mse_test, 2)))print('MSE on train:' + str(np.round(mse_train, 2)))# 分類算法案例# We will create sample data as follows:# x-data: sample 50 random values from a normal = N(-1, 1)#     + sample 50 random values from a normal = N(1, 1)# target: 50 values of 0 + 50 values of 1.#     These are essentially 100 values of the corresponding output index# We will fit the binary classification model:# If sigmoid(x+A) < 0.5 -> 0 else 1# Theoretically, A should be -(mean1 + mean2)/2# 重置計(jì)算圖ops.reset_default_graph()# 加載計(jì)算圖sess = tf.Session()# 聲明批量大小batch_size = 25# 創(chuàng)建數(shù)據(jù)集x_vals = np.concatenate((np.random.normal(-1, 1, 50), np.random.normal(2, 1, 50)))y_vals = np.concatenate((np.repeat(0., 50), np.repeat(1., 50)))x_data = tf.placeholder(shape=[1, None], dtype=tf.float32)y_target = tf.placeholder(shape=[1, None], dtype=tf.float32)# 分割數(shù)據(jù)集 train/test = 80%/20%train_indices = np.random.choice(len(x_vals), round(len(x_vals)*0.8), replace=False)test_indices = np.array(list(set(range(len(x_vals))) - set(train_indices)))x_vals_train = x_vals[train_indices]x_vals_test = x_vals[test_indices]y_vals_train = y_vals[train_indices]y_vals_test = y_vals[test_indices]# 創(chuàng)建變量 (one model parameter = A)A = tf.Variable(tf.random_normal(mean=10, shape=[1]))# Add operation to graph# Want to create the operstion sigmoid(x + A)# Note, the sigmoid() part is in the loss functionmy_output = tf.add(x_data, A)# 增加分類損失函數(shù) (cross entropy)xentropy = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=my_output, labels=y_target))# Create Optimizermy_opt = tf.train.GradientDescentOptimizer(0.05)train_step = my_opt.minimize(xentropy)# Initialize variablesinit = tf.global_variables_initializer()sess.run(init)# 運(yùn)行迭代for i in range(1800):  rand_index = np.random.choice(len(x_vals_train), size=batch_size)  rand_x = [x_vals_train[rand_index]]  rand_y = [y_vals_train[rand_index]]  sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})  if (i+1)%200==0:    print('Step #' + str(i+1) + ' A = ' + str(sess.run(A)))    print('Loss = ' + str(sess.run(xentropy, feed_dict={x_data: rand_x, y_target: rand_y})))# 評(píng)估預(yù)測(cè)# 用squeeze（）函數(shù)封裝預(yù)測(cè)操作，使得預(yù)測(cè)值和目標(biāo)值有相同的維度。y_prediction = tf.squeeze(tf.round(tf.nn.sigmoid(tf.add(x_data, A))))# 用equal（）函數(shù)檢測(cè)是否相等,# 把得到的true或false的boolean型張量轉(zhuǎn)化成float32型，# 再對(duì)其取平均值，得到一個(gè)準(zhǔn)確度值。correct_prediction = tf.equal(y_prediction, y_target)accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))acc_value_test = sess.run(accuracy, feed_dict={x_data: [x_vals_test], y_target: [y_vals_test]})acc_value_train = sess.run(accuracy, feed_dict={x_data: [x_vals_train], y_target: [y_vals_train]})print('Accuracy on train set: ' + str(acc_value_train))print('Accuracy on test set: ' + str(acc_value_test))# 繪制分類結(jié)果A_result = -sess.run(A)bins = np.linspace(-5, 5, 50)plt.hist(x_vals[0:50], bins, alpha=0.5, label='N(-1,1)', color='white')plt.hist(x_vals[50:100], bins[0:50], alpha=0.5, label='N(2,1)', color='red')plt.plot((A_result, A_result), (0, 8), 'k--', linewidth=3, label='A = '+ str(np.round(A_result, 2)))plt.legend(loc='upper right')plt.title('Binary Classifier, Accuracy=' + str(np.round(acc_value_test, 2)))plt.show()