[深度学习1] Logistic regression

[深度学习1] Logistic regression

Logistic regression

公式:
$$y_p=\sigma (w^{T}x+b)$$

sigmoid函数:
$$\sigma (z)=\frac{1}{1+e^{-z}}$$

sigmoid函数的导数:
$$d\sigma (z)/dz=\sigma (z)(1-\sigma (z))$$

损失函数

单个训练样本的损失函数:
$$L(y_p,y)=-ylog(y_p)-(1-y)log(1-y_p)$$
m个训练样本的损失函数:
$$J(w,b)=\frac{1}{m}\sum _{i=1}^m(-y^{(i)}log(y_p^{(i)})-(1-y^{(i)})log(1-y_p^{(i)}))$$

梯度下降

参数更新:
$$w=w-\alpha \frac{\partial J(w,b)}{\partial (w)}$$

$$b=b-\alpha \frac{\partial J(w,b)}{\partial (b)}$$

在Logistic regression中对dJ/dw 和 dJ/db 进行计算:

[公式中的 J和 L都表示损失函数]

根据链式求导法则:
$$\frac{dL}{dz}=\frac{dL}{da}\ast \frac{da}{dz}$$
分别对两部分进行计算
$$\frac{dL}{da}=-y/a+(1-y)/(1-a)$$
(对数函数求导)
$$\frac{da}{dz}=a\ast (1-a)$$
(sigmoid函数求导)

将两部分相乘:
$$dz=\frac{dL}{dz}=(-\frac{y}{a}+\frac{1-y}{1-a})\ast a(1-a)=a-y$$
只需要记住
$$dz=a-y$$
如果一次使用m个样本进行参数更新:
$$d{w}_1=\frac{1}{m}\sum _{i=1}^m{x}_1^{(i)}d{z}^{(i)}=\frac{1}{m}\sum _{i=1}^m{x}_1^{(i)}(a^{(i)}-y^{(i)})$$

$$d{w}_2=\frac{1}{m}\sum _{i=1}^m{x}_2^{(i)}d{z}^{(i)}=\frac{1}{m}\sum _{i=1}^m{x}_2^{(i)}(a^{(i)}-y^{(i)})$$

对w的各个参数的更新按照上式进行,这个公式的由来可以理解如下(这一下部分没找到参考资料, 但应该正确):
$$d{w}_1=\frac{dL}{d{w}_1}$$
接下来就是应用链式求导法则:
$$d{w}_1=\frac{dL}{d{w}_1}=\frac{dL}{dz}\ast \frac{dz}{d{w}_1}$$
而根据
$$z=w^{T}x+b=w_1{x}_1+w_2{x}_2+...+b$$

$$\frac{dz}{d{w}_1}=x_1$$

接下来对b进行更新:
$$db=\frac{1}{m}\sum _{i=1}^{m}d{z}^{(i)}=\frac{1}{m}\sum _{i=1}^m(a^{(i)}-y^{(i)})$$

向量化

如果使用 for 循环完成m个样本的参数更新, 计算效率会非常低, 所以需要进行向量化,提高程序运行速度。
$$Z=w^{T}x+b=np.dot(w.T,x)+b$$

$$A=\sigma (z)$$

$$dz=A-Y$$

$$dw=\frac{1}{m}\ast X\ast d{z}^T$$

$$db=\frac{1}{m}\ast np.sum(dZ,axis=1)$$
因为每个输入x都是一列, 所以 Z 和 A 都是一行, 需要按 axis=1 相加
$$w=w-\alpha \frac{\partial J(w,b)}{\partial (w)}$$

$$b=b-\alpha \frac{\partial J(w,b)}{\partial (b)}$$

编程实现

import numpy as np
import matplotlib.pyplot as plt
import h5py
import scipy
from scipy import ndimage
from lr_utils import load_dataset

#加载数据的函数，数据一斤存放在datasets文件夹下
def load_dataset():
    train_dataset = h5py.File('datasets/train_catvnoncat.h5', "r")
    train_set_x_orig = np.array(train_dataset["train_set_x"][:])  #训练集特征
    train_set_y_orig = np.array(train_dataset["train_set_y"][:])  #训练集标签

    test_dataset = h5py.File('datasets/test_catvnoncat.h5', "r")
    test_set_x_orig = np.array(test_dataset["test_set_x"][:])  # 测试集特征
    test_set_y_orig = np.array(test_dataset["test_set_y"][:])  # 测试集标签

    classes = np.array(test_dataset["list_classes"][:])  # 分类的类别

    train_set_y_orig = train_set_y_orig.reshape((1, train_set_y_orig.shape[0]))
    test_set_y_orig = test_set_y_orig.reshape((1, test_set_y_orig.shape[0]))

    return train_set_x_orig, train_set_y_orig, test_set_x_orig, test_set_y_orig, classes


# 通过上述函数加载数据
train_set_x_orig, train_set_y, test_set_x_orig, test_set_y, classes = load_dataset()
m_train = train_set_x_orig.shape[0]  #训练集样本个数
m_test = test_set_x_orig.shape[0]    #测试机样本个数
num_px = train_set_x_orig.shape[1]   #每张图片的长或者高（因为给定的数据集是正方形的）

#将训练集和测试集的每张图片转为一列
train_set_x_flatten = train_set_x_orig.reshape(train_set_x_orig.shape[0], -1).T
test_set_x_flatten = test_set_x_orig.reshape(test_set_x_orig.shape[0], -1).T

#因为图片每个像素点的最大值为255，通过以下方式将数据进行归一化
train_set_x = train_set_x_flatten/255.
test_set_x = test_set_x_flatten/255.

#定义sigmoid函数
def sigmoid(z):
    s = 1 / (1 + np.exp(-z))
    return s

# 将w和b 进行归一化，w是一列，维度会和每个flatten后的样本大小相同
def initialize_with_zeros(dim):
    w = np.zeros((dim, 1))
    b = 0
    return w, b


# GRADED FUNCTION: propagate

def propagate(w, b, X, Y):
    """
    进行前向传播，并返回dw db和cost
    """

    m = X.shape[1]
    A = sigmoid(np.dot(w.T, X) + b)  # 前向传播
    cost = -1 / m * np.sum(Y * np.log(A) + (1 - Y) * np.log(1 - A))  # 损失函数

    #计算反向传播的dw，db，
    #dz=(A-Y) 当激活函数为sigmoid函数时成立
    dw = 1 / m * np.dot(X, (A - Y).T)
    db = 1 / m * np.sum(A - Y)

    grads = {"dw": dw,
             "db": db}

    return grads, cost

# 参数优化
def optimize(w, b, X, Y, num_iterations, learning_rate, print_cost=False):

    costs = []  #存放cost 的列表，用于绘制损失函数曲线

    for i in range(num_iterations):
        grads, cost = propagate(w, b, X, Y)
        dw = grads["dw"]
        db = grads["db"]

        # 参数更新
        w = w - learning_rate * dw
        b = b - learning_rate * db
        if i % 100 == 0:
            costs.append(cost)
        if print_cost and i % 100 == 0:
            print("Cost after iteration %i: %f" % (i, cost))

    params = {"w": w,
              "b": b}

    grads = {"dw": dw,
             "db": db}

    return params, grads, costs

# 预测
def predict(w, b, X):

    m = X.shape[1]
    Y_prediction = np.zeros((1, m))
    w = w.reshape(X.shape[0], 1)
    A = sigmoid(np.dot(w.T, X) + b)

    for i in range(A.shape[1]):
        if A[0, i] <= 0.5:
            Y_prediction[0, i] = 0
        else:
            Y_prediction[0, i] = 1

    return Y_prediction


# 调用上述函数，组成一个统一的模型
def model(X_train, Y_train, X_test, Y_test, num_iterations=2000, learning_rate=0.5, print_cost=False):

    w, b = initialize_with_zeros(X_train.shape[0])

    parameters, grads, costs = optimize(w, b, X_train, Y_train, num_iterations, learning_rate, print_cost)

    w = parameters["w"]
    b = parameters["b"]

    Y_prediction_test = predict(w, b, X_test)
    Y_prediction_train = predict(w, b, X_train)


    print("train accuracy: {} %".format(100 - np.mean(np.abs(Y_prediction_train - Y_train)) * 100))
    print("test accuracy: {} %".format(100 - np.mean(np.abs(Y_prediction_test - Y_test)) * 100))

    d = {"costs": costs,
         "Y_prediction_test": Y_prediction_test,
         "Y_prediction_train": Y_prediction_train,
         "w": w,
         "b": b,
         "learning_rate": learning_rate,
         "num_iterations": num_iterations}

    return d

d = model(train_set_x, train_set_y, test_set_x, test_set_y, num_iterations = 2000, learning_rate = 0.005, print_cost = True)

# 绘制损失函数变化曲线
costs = np.squeeze(d['costs'])
plt.figure()
plt.plot(costs)
print(costs)
plt.ylabel('cost')
plt.xlabel('iterations (per hundreds)')
plt.title("Learning rate =" + str(d["learning_rate"]))
plt.show()

参考资料

1. 吴恩达深度学习教程