Add exam.9

exam/step9/outputsUtilsCompleted.py

+import numpy as np
+
+def softmax(x):
+    np.seterr(divide='ignore', invalid='ignore')
+    return (np.exp(x).T / np.sum(np.exp(x), axis=1)).T
+
+
+def returnOneHot(Output):
+    '''
+    :param Output: 神经网络的输出
+    :return:
+    '''
+    out = np.zeros(Output.shape,dtype=np.int)
+    idx = Output.argmax(axis=1)
+    out[np.arange(Output.shape[0]), idx] = 1
+    return out
+
+def computeAccuracy(pred,label):
+    '''
+    :param pred: 预测值
+    :param label: 实际值
+    :return:
+    '''
+    right = 0
+    for p,l in zip(pred,label):
+        if (p==l).all():
+            right+=1
+    return right/len(pred)
\ No newline at end of file

exam/step9/outputsUtilsForUsers.py

+import numpy as np
+
+def softmax(x):
+    '''
+    对输出的每一行(即对每个样本的各个标签上的值)做softmax变换
+    :param x: 一般是(batchSize,num_Labels), 是模型的输出
+    :return: 每一行经过了softmax处理之后的结果
+    '''
+
+    #********** Begin **********#
+
+    #********** End **********#
+
+
+def returnOneHot(Output):
+    '''
+    softmax的输出不是onehot型的, 将最大概率处替换为1，其他位置均置为0
+    :param Output: 神经网络的输出
+    :return: onehot型的输出
+    '''
+
+    #********** Begin **********#
+
+    #********** End **********#
+
+
+def computeAccuracy(pred,label):
+    '''
+    计算预测的正确率
+    :param pred: 预测的标签
+    :param label:  预测样本真实的标签
+    :return: 正确率
+    '''
+
+    #********** Begin **********#
+
+    #********** End **********#
+

exam/step9/outputsUtilsTest.py

+from outputsUtilsForUsers import softmax as Softmax
+from outputsUtilsForUsers import computeAccuracy as ComputeAccuracy
+from outputsUtilsForUsers import returnOneHot as ReturnOneHot
+#
+# from outputsUtilsForUsers import softmax as Softmax
+# from outputsUtilsForUsers import computeAccuracy as ComputeAccuracy
+# from outputsUtilsForUsers import returnOneHot as ReturnOneHot
+
+
+'''
+测试矩阵:
+[[-1  2 25  7]
+ [20 15 10  5]
+ [12  4 19  5]
+ [ 9 13 61  8]]
+对应标签:
+[[0 0 1 0]
+ [1 0 0 0]
+ [0 0 1 0]
+ [0 0 1 0]]
+你softmax的值:
+[[0.    0.    1.    0.   ]
+ [0.993 0.007 0.    0.   ]
+ [0.001 0.    0.999 0.   ]
+ [0.    0.    1.    0.   ]]
+你onehot编码后的值:
+[[0 0 1 0]
+ [1 0 0 0]
+ [0 0 1 0]
+ [0 0 1 0]]
+你计算的准确率:
+1.0
+
+'''
+
+
+import numpy as np
+
+test_matrix = np.array([[-1,2,25,7],
+                        [20,15,10,5],
+                        [12,4,19,5],
+                        [9,13,61,8],])
+
+test_label = np.array([[0,0,1,0],
+                       [1,0,0,0],
+                       [0,0,1,0],
+                       [0,0,1,0]])
+
+
+print('测试矩阵:')
+print(test_matrix)
+print('对应标签:')
+print(test_label)
+
+softmaxed = Softmax(test_matrix)
+print('你softmax的值:')
+print(np.around(softmaxed,decimals=3))
+
+onehoted = ReturnOneHot(softmaxed)
+print('你onehot编码后的值:')
+print(onehoted)
+
+accuracy = ComputeAccuracy(onehoted,test_label)
+print('你计算的准确率:')
+print(accuracy)