损失函数在机器学习中的模型非常重要的一部分,它代表了评价模型的好坏程度的标准,最终的优化目标就是通过调整参数去使得损失函数尽可能的小,如果损失函数定义错误或者不符合实际意义的话,训练模型只是在浪费时间。
所以先来了解一下常用的几个损失函数hinge loss(合页损失)、softmax loss、cross_entropy loss(交叉熵损失):
1:hinge loss(合页损失)
又叫Multiclass SVM loss。至于为什么叫合页或者折页函数,可能是因为函数图像的缘故。
s=WX,表示最后一层的输出,维度为(C,None),$L_i$表示每一类的损失,一个样例的损失是所有类的损失的总和。
$L_i=\sum_{j!=y_i}\left \{ ^{0 \ \ \ \ \ \ \ \ if \ s_{y_i}\geq s_j+1}_{s_j-s_{y_i}+1 \ otherwise} \right \}$
$ =\sum_{j!=y_i}max(0,s_{y_i}-s_j+1)$
函数图像长这样:
举个例子:
假设我们只有3个类别,上面的图片表示输入,下面3个数字表示最后一层每个类别的分数。
对于第一张猫的图片,$L_i$=max(0, 5.1 - 3.2 + 1) +max(0, -1.7 - 3.2 + 1)=2.9+0=2.9
对于第二张汽车的图片,$L_i$=max(0, 1.3 - 4.9 + 1) +max(0, 2.0 - 4.9 + 1)=0+0=0
可以看到对于分类错误的样例,最后有一个损失产生,对于正确分类的样例,其损失为0.
其实该损失函数不仅仅只是要求正确类别的分数最高,还要高出一定程度。也就是说即使分类正确了也有可能会产生损失,因为有可能正确的类别的分数没有超过错误类别一定的阈值(这里设为1)。但是不管阈值设多少,好像并没有什么太大的意义,只是对应着W的放缩而已。简单的理解就是正确类别的得分不仅要最高,而且要高的比较明显。
对于随机初始化的权重,最终输出应该也不叫均匀,loss应该能得到C-1,可以用这一点来检验自己的损失函数和前向传播的实现是否正确。
2:softmax
softmax操作的意图是把分数转换成概率,来看看怎么转换.
对于输入$x_i$最后会得到C个类别的分数s,每个类别的概率,$P(Y=k|X=x_i)=\frac{e^{s_k}}{\sum_j e^{s_j}}$。
首先取指数次幂,得到整数,然后用比值代替概率。
这样转换了之后我们定义似然函数(也就是损失函数)$L_i=-logP(Y=y_i|X=x_i)$,也就是说正确类别的概率越大越好,这也很好理解。
同样的我们来看一个例子:
同样的,对于随机初始话的权重和数据,$L_i$应该=-log(1/c)。
3:cross_entrop loss
熵的本质是香农信息量的期望。
现有关于样本集的2个概率分布p和q,其中p为真实分布,q非真实分布。按照真实分布p来衡量识别一个样本的所需要的编码长度的期望(即平均编码长度)为:H(p)=-∑p(i)∗logp(i),如果使用错误分布q来表示来自真实分布p的平均编码长度,则应该是:H(p,q)=-∑p(i)∗logq(i)。H(p,q)=-∑p(i)∗logq(i)。因为用q来编码的样本来自分布p,所以期望H(p,q)中概率是p(i)。H(p,q)我们称之为“交叉熵”。
根据公式可以看出,对于正确分类只有一个情况,交叉熵H(p,q)=- ∑logq(i),其中q(i)为正确分类的预测概率,其余的项全部为0,因为除了正确类别,其余的真是概率p(i)都为0.在这种情况下,交叉熵损失与softmax损失其实就是同一回事。
python代码实现:
1 #首先是线性分类器的类实现 linear_classifier.py
2
3 importnumpy as np4 from linear_svm import *
5 from softmax import *
6
7
8 classLinearClassifier(object):9 #线性分类器的基类
10 def __init__(self):11 self.W =None12
13 def train(self, X, y, learning_rate=1e-3, reg=1e-5, num_iters=100,14 batch_size=200, verbose=False):15 """
16 使用SGD优化参数矩阵17
18 Inputs:19 - X (N, D)20 - y (N,)21 - learning_rate: 学习率.22 - reg: 正则参数.23 - num_iters: (int) 训练迭代的次数24 - batch_size: (int) 每次迭代使用的样本数量.25 - verbose: (boolean) 是否显示训练进度26
27 Outputs:28 返回一个list保存了每次迭代的loss29 """
30 num_train, dim =X.shape31 num_classes = np.max(y) + 1 #假设有k类,y的取值为【0,k-1】且最大的下标一定会在训练数据中出现
32
33 #初始化权重矩阵
34 if self.W isNone:35 self.W = 0.001 *np.random.randn(dim, num_classes)36
37
38 loss_history =[]39 for it inrange(num_iters):40 #在每次迭代,随机选择batch_size个数据
41 mask=np.random.choice(num_train,batch_size,replace=True)42 X_batch =X[mask]43 y_batch =y[mask]44
45 #计算损失和梯度
46 loss, grad =self.loss(X_batch, y_batch, reg)47 loss_history.append(loss)48
49 #更新参数
50 self.W-=grad*learning_rate51
52 if verbose and it % 100 ==0:53 print('iteration %d / %d: loss %f' %(it, num_iters, loss))54
55 returnloss_history56
57 defpredict(self, X):58 """
59 使用训练好的参数来对输入进行预测60
61 Inputs:62 - X (N, D)63
64 Returns:65 - y_pred (N,):预测的正确分类的下标66 """
67
68 y_pred=np.dot(X,self.W)69 y_pred = np.argmax(y_pred, axis = 1)70
71 returny_pred72
73 defloss(self, X_batch, y_batch, reg):74 """
75 这只是一个线性分类器的基类76 不同的线性分类器loss的计算方式不同77 所以需要在子类中重写78 """
79 pass
80
81
82 classLinearSVM(LinearClassifier):83 """使用SVM loss"""
84
85 defloss(self, X_batch, y_batch, reg):86 returnsvm_loss_vectorized(self.W, X_batch, y_batch, reg)87
88
89 classSoftmax(LinearClassifier):90 """使用交叉熵"""
91
92 defloss(self, X_batch, y_batch, reg):93 return softmax_loss_vectorized(self.W, X_batch, y_batch, reg)
1 #svm loss 的实现 softmax.py
2
3 importnumpy as np4 from random importshuffle5
6 defsoftmax_loss_naive(W, X, y, reg):7 """
8 用循环实现softmax损失函数9 D,C,N分别表示数据维度,标签种类个数和数据批大小10 Inputs:11 - W (D, C):weights.12 - X (N, D):data.13 - y (N,): labels14 - reg: (float) regularization strength15
16 Returns :17 - loss18 - gradient19 """
20
21 loss = 0.0
22 dW =np.zeros_like(W)23
24 num_classes = W.shape[1]25 num_train =X.shape[0]26
27 for i inrange(num_train):28 scores=np.dot(X[i],W)29 shift_scores=scores-max(scores)30 dom=np.log(np.sum(np.exp(shift_scores)))31 loss_i=-shift_scores[y[i]]+dom32 loss+=loss_i33 for j inrange(num_classes):34 softmax_output = np.exp(shift_scores[j])/sum(np.exp(shift_scores))35 if j ==y[i]:36 dW[:,j] += (-1 + softmax_output) *X[i].T37 else:38 dW[:,j] += softmax_output *X[i].T39 loss /=num_train40 loss += reg * np.sum(W *W)41 dW = dW/num_train + 2*reg*W42
43
44 returnloss, dW45
46
47 defsoftmax_loss_vectorized(W, X, y, reg):48 """
49 无循环的实现50 """
51
52 loss = 0.0
53 dW =np.zeros_like(W)54 num_classes = W.shape[1]55 num_train =X.shape[0]56
57 scores=np.dot(X,W)58 shift_scores=scores-np.max(scores,axis=1).reshape(-1,1)59 softmax_output = np.exp(shift_scores)/np.sum(np.exp(shift_scores), axis = 1).reshape(-1,1)60 loss=np.sum(-np.log(softmax_output[range(num_train),y]))61 loss=loss/num_train+reg * np.sum(W *W)62
63 dW=softmax_output.copy()64 dW[range(num_train),y]-=1
65 dW=np.dot(X.T,dW)66 dW = dW/num_train + 2*reg*W67
68
69 return loss, dW
1 #svm loss的实现 linear_svm.py
2
3 importnumpy as np4 from random importshuffle5
6 defsvm_loss_naive(W, X, y, reg):7 """
8 用循环实现的SVM loss计算9 这里的loss函数使用的是margin loss10
11 Inputs:12 - W (D, C): 权重矩阵.13 - X (N, D): 批输入14 - y (N,) 标签15 - reg: 正则参数16
17 Returns :18 - loss float19 - W的梯度20 """
21 dW =np.zeros(W.shape)22 num_classes = W.shape[1]23 num_train =X.shape[0]24 loss = 0.0
25
26 for i inrange(num_train):27 scores =X[i].dot(W)28 correct_class_score =scores[y[i]]29 for j inrange(num_classes):30 if j ==y[i]:31 continue
32 margin = scores[j] - correct_class_score + 1
33 if margin >0:34 loss +=margin35 dW[:,j]+=X[i].T36 dW[:,y[i]]-=X[i].T37
38
39 loss /=num_train40 dW/=num_train41 loss += reg * np.sum(W *W)42 dW+=2* reg *W43
44
45 returnloss, dW46
47
48 defsvm_loss_vectorized(W, X, y, reg):49 """
50 不使用循环,利用numpy矩阵运算的特性实现loss和梯度计算51 """
52 loss = 0.0
53 dW =np.zeros(W.shape)54
55 #计算loss
56 num_classes = W.shape[1]57 num_train =X.shape[0]58 scores=np.dot(X,W)#得到得分矩阵(N,C)
59 correct_socre=scores[range(num_train), list(y)].reshape(-1,1)#得到每个输入的正确分类的分数
60 margins=np.maximum(0,scores-correct_socre+1)61 margins[range(num_train), list(y)] =062 loss=np.sum(margins)/num_train+reg * np.sum(W *W)63
64 #计算梯度
65 mask=np.zeros((num_train,num_classes))66 mask[margins>0]=1
67 mask[range(num_train),list(y)]-=np.sum(mask,axis=1)68 dW=np.dot(X.T,mask)69 dW/=num_train70 dW+=2* reg *W71
72 return loss, dW
1 #最后是测试文件,看看实现的线性分类器在CIFAR10上的分类效果如何
2
3 #coding: utf-8
4
5 #实现hinge_loss和sotfmax_loss
6
7 importrandom8 importnumpy as np9 from cs231n.data_utils importload_CIFAR1010 importmatplotlib.pyplot as plt11 from cs231n.classifiers.linear_svm importsvm_loss_naive,svm_loss_vectorized12 from cs231n.classifiers.softmax importsoftmax_loss_naive,softmax_loss_vectorized13 importtime14 from cs231n.classifiers importLinearSVM,Softmax15 %matplotlib inline16 plt.rcParams['figure.figsize'] = (10.0, 8.0) #set default size of plots
17 plt.rcParams['image.interpolation'] = 'nearest'
18 plt.rcParams['image.cmap'] = 'gray'
19
20
21 #######################################################################################
22 #######################################################################################
23 ###################################第一部分 载入数据并处理###############################
24 #######################################################################################
25 #######################################################################################
26
27 #载入CIFAR10数据.
28 cifar10_dir = 'cs231n/datasets/cifar-10-batches-py'
29 X_train, y_train, X_test, y_test =load_CIFAR10(cifar10_dir)30
31 print('Training data shape:', X_train.shape)32 print('Training labels shape:', y_train.shape)33 print('Test data shape:', X_test.shape)34 print('Test labels shape:', y_test.shape)35
36 #每个分类选几个图片显示观察一下
37 classes = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']38 num_classes =len(classes)39 samples_per_class = 7
40 for y, cls inenumerate(classes):41 idxs = np.flatnonzero(y_train ==y)42 idxs = np.random.choice(idxs, samples_per_class, replace=False)43 for i, idx inenumerate(idxs):44 plt_idx = i * num_classes + y + 1
45 plt.subplot(samples_per_class, num_classes, plt_idx)46 plt.imshow(X_train[idx].astype('uint8'))47 plt.axis('off')48 if i ==0:49 plt.title(cls)50 plt.show()51
52 #把数据分为训练集,验证集和测试集。
53 #用一个小子集做测验,运行更快。
54 num_training = 49000
55 num_validation = 1000
56 num_test = 1000
57 num_dev = 500
58
59 #数据集本身没有给验证集,需要自己把训练集分成两部分
60 mask = range(num_training, num_training +num_validation)61 X_val =X_train[mask]62 y_val =y_train[mask]63
64 mask =range(num_training)65 X_train =X_train[mask]66 y_train =y_train[mask]67
68
69 mask = np.random.choice(num_training, num_dev, replace=False)70 X_dev =X_train[mask]71 y_dev =y_train[mask]72
73 mask =range(num_test)74 X_test =X_test[mask]75 y_test =y_test[mask]76
77 print('Train data shape:', X_train.shape)78 print('Train labels shape:', y_train.shape)79 print('Validation data shape:', X_val.shape)80 print('Validation labels shape:', y_val.shape)81 print('Test data shape:', X_test.shape)82 print('Test labels shape:', y_test.shape)83
84
85 X_train = np.reshape(X_train, (X_train.shape[0], -1))86 X_val = np.reshape(X_val, (X_val.shape[0], -1))87 X_test = np.reshape(X_test, (X_test.shape[0], -1))88 X_dev = np.reshape(X_dev, (X_dev.shape[0], -1))89
90 print('Training data shape:', X_train.shape)91 print('Validation data shape:', X_val.shape)92 print('Test data shape:', X_test.shape)93 print('dev data shape:', X_dev.shape)94
95
96 #预处理: 把像素点数据化成以0为中心
97 #第一步: 在训练集上计算图片像素点的平均值
98 mean_image = np.mean(X_train, axis=0)99 print(mean_image.shape)100 plt.figure(figsize=(4,4))101 plt.imshow(mean_image.reshape((32,32,3)).astype('uint8')) #可视化一下平均值
102 plt.show()103
104 #第二步: 所有数据都减去刚刚得到的均值
105 X_train -=mean_image106 X_val -=mean_image107 X_test -=mean_image108 X_dev -=mean_image109
110
111 #第三步: 给所有的图片都加一个位,并设为1,这样在训练权重的时候就不需要b了,只需要w
112 #相当于把b的训练并入了W中
113 X_train = np.hstack([X_train, np.ones((X_train.shape[0], 1))])114 X_val = np.hstack([X_val, np.ones((X_val.shape[0], 1))])115 X_test = np.hstack([X_test, np.ones((X_test.shape[0], 1))])116 X_dev = np.hstack([X_dev, np.ones((X_dev.shape[0], 1))])117
118 print(X_train.shape, X_val.shape, X_test.shape, X_dev.shape)119
120
121 #######################################################################################
122 #######################################################################################
123 ###################################第二部分 定义需要用到的函数###########################
124 #######################################################################################
125 #######################################################################################
126
127 defcmp_naiveANDvectorized(naive,vectorized):128 '''
129 每个损失函数都用两种方式实现:循环和无循环(即利用numpy的特性)130 '''
131
132 W = np.random.randn(3073, 10) * 0.0001
133
134 #对比两张实现方式的计算时间
135 tic =time.time()136 loss_naive, grad_naive = naive(W, X_dev, y_dev, 0.000005)137 toc =time.time()138 print('Naive computed in %fs' % ( toc -tic))139
140 tic =time.time()141 loss_vectorized, grad_vectorized = vectorized(W, X_dev, y_dev, 0.000005)142 toc =time.time()143 print('Vectorized computed in %fs' % ( toc -tic))144
145 #检验损失的实现是否正确,对于随机初始化的数据的权重,
146 #softmax_loss应该约等于-log(0.1),svm_loss应该约等于9
147 print('loss %f %f' %(loss_naive , loss_vectorized))148
149 #对比两种实现方式得到的结果是否相同
150 print('difference loss %f' % (loss_naive -loss_vectorized))151 difference = np.linalg.norm(grad_naive - grad_vectorized, ord='fro')152 print('difference gradient: %f' %difference)153
154 defcross_choose(Linear_classifier,learning_rates,regularization_strengths):155 '''
156 选择超参数157 '''
158
159 results = {} #存储每一对超参数对应的训练集和验证集上的正确率
160 best_val = -1 #最好的验证集上的正确率
161 best_model = None #最好的验证集正确率对应的svm类的对象
162 best_loss_hist=None163 for rs inregularization_strengths:164 for lr inlearning_rates:165 classifier =Linear_classifier166 loss_hist = classifier.train(X_train, y_train, lr, rs, num_iters=3000)167 y_train_pred =classifier.predict(X_train)168 train_accuracy = np.mean(y_train ==y_train_pred)169 y_val_pred =classifier.predict(X_val)170 val_accuracy = np.mean(y_val ==y_val_pred)171 if val_accuracy >best_val:172 best_val =val_accuracy173 best_model =classifier174 best_loss_hist=loss_hist175 results[(lr,rs)] =train_accuracy, val_accuracy176
177 for lr, reg insorted(results):178 train_accuracy, val_accuracy =results[(lr, reg)]179 print ('lr %e reg %e train accuracy: %f val accuracy: %f' %(180 lr, reg, train_accuracy, val_accuracy))181
182 print('best validation accuracy achieved during cross-validation: %f' %best_val)183 #可视化loss曲线
184 plt.plot(best_loss_hist)185 plt.xlabel('Iteration number')186 plt.ylabel('Loss value')187 plt.show()188 returnresults,best_val,best_model189
190 defshow_weight(best_model):191 #看看最好的模型的效果
192 #可视化学到的权重
193
194 y_test_pred =best_model.predict(X_test)195 test_accuracy = np.mean(y_test ==y_test_pred)196 print('final test set accuracy: %f' %test_accuracy)197
198 w = best_model.W[:-1,:] #去掉偏置参数
199 w = w.reshape(32, 32, 3, 10)200 w_min, w_max =np.min(w), np.max(w)201 classes = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']202 for i in range(10):203 plt.subplot(2, 5, i + 1)204
205 #把权重转换到0-255
206 wimg = 255.0 * (w[:, :, :, i].squeeze() - w_min) / (w_max -w_min)207 plt.imshow(wimg.astype('uint8'))208 plt.axis('off')209 plt.title(classes[i])210 plt.show()211
212 #######################################################################################
213 #######################################################################################
214 ##########################第三部分 应用和比较svm_loss和softmax_loss######################
215 #######################################################################################
216 #######################################################################################
217
218 cmp_naiveANDvectorized(svm_loss_naive,svm_loss_vectorized)219 learning_rates = [(1+i*0.1)*1e-7 for i in range(-3,5)]220 regularization_strengths = [(5+i*0.1)*1e3 for i in range(-3,3)]221 #正则参数的选择要根据正常损失和W*W的大小的数量级来确定,初始时正常loss大概是9,W*W大概是1e-6
222 #可以观察最后loss的值的大小来继续调整正则参数的大小,使正常损失和正则损失保持合适的比例
223 results,best_val,best_model=cross_choose(LinearSVM(),learning_rates,regularization_strengths)224 show_weight(best_model)225
226 print("--------------------------------------------------------")227
228 cmp_naiveANDvectorized(softmax_loss_naive,softmax_loss_vectorized)229 learning_rates = [(2+i*0.1)*1e-7 for i in range(-2,2)]230 regularization_strengths = [(7+i*0.1)*1e3 for i in range(-3,3)]231 results,best_val,best_model=cross_choose(Softmax(),learning_rates,regularization_strengths)232 show_weight(best_model)
1 #获取数据的部分
2
3 from six.moves importcPickle as pickle4 importnumpy as np5 importos6 from scipy.misc importimread7 importplatform8
9 defload_pickle(f):10 version =platform.python_version_tuple()11 if version[0] == '2':12 returnpickle.load(f)13 elif version[0] == '3':14 return pickle.load(f, encoding='latin1')15 raise ValueError("invalid python version: {}".format(version))16
17 defload_CIFAR_batch(filename):18 """CIRAR的数据是分批的,这个函数的功能是载入一批数据"""
19 with open(filename, 'rb') as f:20 datadict = load_pickle(f) #以二进制方式打开文件
21 X = datadict['data']22 Y = datadict['labels']23 X = X.reshape(10000, 3, 32, 32).transpose(0,2,3,1).astype("float")24 Y =np.array(Y)25 returnX, Y26
27 defload_CIFAR10(ROOT):28 """load 所有的数据"""
29 xs =[]30 ys =[]31 for b in range(1,6):32 f = os.path.join(ROOT, 'data_batch_%d' %(b, ))33 X, Y =load_CIFAR_batch(f)34 xs.append(X)35 ys.append(Y)36 Xtr =np.concatenate(xs)37 Ytr =np.concatenate(ys)38 delX, Y39 Xte, Yte = load_CIFAR_batch(os.path.join(ROOT, 'test_batch'))40 returnXtr, Ytr, Xte, Yte41
42
43 def get_CIFAR10_data(num_training=49000, num_validation=1000, num_test=1000,44 subtract_mean=True):45 """
46 Load the CIFAR-10 dataset from disk and perform preprocessing to prepare47 it for classifiers. These are the same steps as we used for the SVM, but48 condensed to a single function.49 """
50 #Load the raw CIFAR-10 data
51 cifar10_dir = 'cs231n/datasets/cifar-10-batches-py'
52 X_train, y_train, X_test, y_test =load_CIFAR10(cifar10_dir)53
54 #Subsample the data
55 mask = list(range(num_training, num_training +num_validation))56 X_val =X_train[mask]57 y_val =y_train[mask]58 mask =list(range(num_training))59 X_train =X_train[mask]60 y_train =y_train[mask]61 mask =list(range(num_test))62 X_test =X_test[mask]63 y_test =y_test[mask]64
65 #Normalize the data: subtract the mean image
66 ifsubtract_mean:67 mean_image = np.mean(X_train, axis=0)68 X_train -=mean_image69 X_val -=mean_image70 X_test -=mean_image71
72 #Transpose so that channels come first
73 X_train = X_train.transpose(0, 3, 1, 2).copy()74 X_val = X_val.transpose(0, 3, 1, 2).copy()75 X_test = X_test.transpose(0, 3, 1, 2).copy()76
77 #Package data into a dictionary
78 return{79 'X_train': X_train, 'y_train': y_train,80 'X_val': X_val, 'y_val': y_val,81 'X_test': X_test, 'y_test': y_test,82 }