贴部分代码供参考
  
#data normalization
def normalization(data):
     _range = np.max(data) - np.min(data)
     return (data - np.min(data))/_range
  
# sigmod function
def sigmod(z):
     a = 1/(1+np.exp(-z))
     return a
  
#initialize the parameters
def initialize_with_zeros(dim):
    w = np.zeros((dim,1))
    b = 0
    return w,b
  
def propagate(w,b,X,Y):
     #获取样本数
     m = X.shape[1]
     #前向传播，交叉商
     A = sigmod(np.dot(w.T,X)+b)
     cost = -(np.sum(Y*np.log(A)+(1-Y)*np.log(1-A)))/m
     #反向传播
     dZ = A - Y
     dw =(np.dot(X,dZ.T))/m
     db = (np.sum(dZ))/m
     #返回值
     grads = {"dw":dw,"db":db}
     return grads, cost
  
  
#define the optimize function
def optimize(w,b,X,Y,num_iterations,learning_rate):
     costs = []
     for i in range(num_iterations):
        grads, cost = propagate(w,b,X,Y)
        dw = grads["dw"]
        db = grads["db"]
        #更新参数
        w = w - learning_rate*dw
        b = b - learning_rate*db
        if i%100 ==0:
            costs.append(cost)
     params = {"w":w,"b":b}
     grads = {"dw":dw,"db":db}
     return params,grads,costs
  
def predict(w,b,X):
     m = X.shape[1]
     Y_prediction = np.zeros((1,m))
  
     A = sigmod(np.dot(w.T,X)+b)
     for i in range(m):
         if A[0,i]>0.5:
             Y_prediction[0,i] = 1
         else:
             Y_prediction[0,i] = 0
     return Y_prediction
  
def logistic_model(X_train, Y_train, X_test, Y_test, learning_rate, num_iterations):
     learning_rate = 0.1
     dim = X_train.shape[0]
     W,b = initialize_with_zeros(dim)
     params, grads, costs = optimize(W,b,X_train,Y_train,num_iterations,learning_rate)
     W = params['w']
     b = params['b']
     prediction_train = predict(W,b,X_test)
     prediction_test = predict(W,b,X_train)
     accuracy_train = 1- np.mean(np.abs(prediction_train-Y_train))
     accuracy_test = 1 - np.mean(np.abs(prediction_test-Y_test))
     print("Accuracy on train set:",accuracy_train)
     print("Accuracy on test set:",accuracy_test)
     d = {"costs":costs,
          "Y_prediction_test":prediction_test,
          "Y_prediction_train":prediction_train,
          "w":w,
          "b":b,
          "learning_rate":learning_rate,
          "num_iterations":num_iterations,
     }
     return d
  
【 在 puke 的大作中提到: 】
: 用python全手写？好歹得能用numpy吧？
--
FROM 219.239.227.*