文章内容为吴恩达深度学习第二周的编程作业
**ipynbg格式代码及数据集-->陈能豆 **
密码: o1bn
#!/usr/bin/env python
# coding: utf-8
# # 吴恩达
深度学习第二周编程作业
# ## 题目部分
# 先来看一下 英文的题目部分
# Logistic Regression with a Neural Network mindset
# Welcome to your first (required) programming assignment! You will build a logistic regression classifier to recognize cats. This assignment will step you through how to do this with a Neural Network mindset, and so will also hone your intuitions about deep learning.
#
# Instructions:
#
# Do not use loops (for/while) in your code, unless the instructions explicitly ask you to do so.
# You will learn to:
#
# Build the general architecture of a learning algorithm, including:
# Initializing parameters
# Calculating the cost function and its gradient
# Using an optimization algorithm (gradient descent)
# Gather all three functions above into a main model function, in the right order.
# ### 简单解释一下要求
# 要求我们建议一个逻辑回归分类器来识别猫的图片
#
# #### 要求如下:
# * 不要在代码中使用循环(for/while),除非指令明确要求您这样做。
#
# * 算法的架构如下:
# 1. 参数初始化
# 2. 成本函数及梯度的计算
# 3. 优化算法
#
# 你需要编写上面提到的方法,并进行组合,成为一个模型
#
# #### 数据集概述
# 这里会提供给我们一个数据集,名字为”data.h5“(和下面代码用的是同一个数据集,不过名字改成了”train_catvnoncat.h5“)
# 里面包含了相关数据集,更具体的可以看代码的数据导入方法。
# 主要包括:
# 1. 标签设置好的训练集,标记是否为cat(0/1)
# 2. 标签设置好的测试集
# 3. 图片的形状为(64, 64, 3)
#
#
# 数据集及代码附上链接--> [陈能豆](https://pan.baidu.com/s/1VMCuVU8IKWLycKM6B20O_g)
# 提取密码:q73d
#
# #### ok 下面看代码
#
# In[98]:
# pip --version # 这里看一下python的版本 ,3.6 以及别的版本应该也是可行的
# ### 代码部分
# #### 代码段后面附上相关重点提示
# #### 相关函数用法 不会的百度
# In[197]:
import numpy as np
import matplotlib.pyplot as plt
import h5py
import random
# 下面两个用于测试模型
from PIL import Image
import imageio
# 导入相关包
# In[198]:
# 导入数据函数
def load_datasets():
train_datasets = h5py.File('datasets/train_catvnoncat.h5',"r")
train_set_x_origi = np.array(train_datasets["train_set_x"][:])
train_set_y_origi = np.array(train_datasets["train_set_y"][:])
# 读取训练集
test_datasets = h5py.File('datasets/test_catvnoncat.h5',"r")
test_set_x_origi = np.array(test_datasets["test_set_x"][:])
test_set_y_origi = np.array(test_datasets["test_set_y"][:])
# 测试集
classes = np.array(test_datasets["list_classes"][:])
# 存储可能的结果 0 non-cat 1 cat
# 初始化矩阵格式
train_set_y_origi = train_set_y_origi.reshape( (1,train_set_y_origi.shape[0]) )
test_set_y_origi = test_set_y_origi.reshape( (1,test_set_y_origi.shape[0]) )
return train_set_x_origi,train_set_y_origi,test_set_x_origi,test_set_y_origi,classes
# 从数据集中导入数据 其中classes 用于存储可能的结果
# classes: array([b'non-cat', b'cat'], dtype='|S7')
# In[199]:
# 读入数据
train_set_x_origi,train_set_y,test_set_x_origi,test_set_y,classes = load_datasets()
# In[200]:
# 显示一下训练集的数据
index = random.randint(1,train_set_y.shape[1])
plt.imshow(train_set_x_origi[index])
# 打印图片
# 打印结果
print( "The picture number is " +str(index) + " The ans is " +classes[ np.squeeze(train_set_y[:,index])].decode("utf-8") )
print("使用np.squeeze:" + str(np.squeeze(train_set_y[:,index])) )
print("不使用np.squeeze: " + str(train_set_y[:,index]) )
# 显示的是数据集中随机的一张图片及结果
# 可改为指定编号
# squeeze 函数的作用为压缩维度
# 是否使用squeeze的区别见代码
#
# In[201]:
# 查看矩阵维数
im_number_train = train_set_y.shape[1]
im_number_test = test_set_y.shape[1]
im_size = train_set_x_origi.shape[1]
# 打印
print("训练集图片数量: "+str(im_number_train))
print("测数集图片数量: "+str(im_number_test))
print("图片宽度高度为: "+str(im_size))
print("每张图片大小为" +str(im_size)+","+str(im_size)+",3")
print("训练集x维数为 "+str(train_set_x_origi.shape))
print("训练集y维数为 "+str(train_set_y.shape))
print("测试集x维数为 "+str(test_set_x_origi.shape))
print("测试集y维数为 "+str(test_set_y.shape))
# shaape 函数 返回矩阵维数
# 返回数据类型为n元组 可通过下标引用
# In[202]:
# 降维 数据预处理
train_set_x_flatten = train_set_x_origi.reshape(train_set_x_origi.shape[0],-1).T
test_set_x_flatten = test_set_x_origi.reshape(test_set_x_origi.shape[0],-1).T
# 打印降维后的维数
print("训练集降维后的维数为 "+str(train_set_x_flatten.shape))
print("测试集降维后的维数为 "+str(test_set_x_flatten.shape))
# reshape 函数用于改变矩阵维数
# -1 表示 自动计算下一维度
# In[203]:
# 核对维数
print("训练集x的位数为 "+str(train_set_x_flatten.shape))
print("训练集y维数为 "+str(train_set_y.shape))
print("测试集x的维数为 "+str(test_set_x_flatten.shape))
print("测试集y维数为 "+str(test_set_y.shape))
# In[204]:
# 数据标准化
# 图片存储的值为像素值 0-255
# 例如数据集中
print(np.amax(train_set_x_flatten))
print(np.amin(train_set_x_flatten))
train_set_x = train_set_x_flatten/255
test_set_x = test_set_x_flatten/255
# In[205]:
def sigmoid(z):
# 参数为z数组
# 返回s = sigmoid(z)
s = 1/(1+np.exp(-z))
return s
# 激活函数有很多种
# ![image.png](attachment:image.png "激活函数")
# In[206]:
# 测试siggmod 函数 是否符合预期
print( "sigmoid(0) = "+str(sigmoid(0)))
print( "sigmoid(1) = "+str(sigmoid(1)))
print( "sigmoid(10) = "+str(sigmoid(10)))
print( "sigmoid(-1) = "+str(sigmoid(-1)))
print( "sigmoid(-10) = "+str(sigmoid(-10)))
# In[207]:
def initialize_zeros_vector(size):
# 函数作用 生成 并返回一个(size,1) 的0向量 同时初始化 b=0
w = np.zeros( (size,1) )
b = 0
# 断言确保维数
assert( w.shape== (size,1) )
assert( isinstance(b,float) or isinstance(b,int) )
return (w , b)
# In[208]:
def two_way_propogate(W,b,X,Y ):
# 实现正向传播和反向传播
m = X.shape[1]
# 正向传播
A = sigmoid(np.dot(W.T,X )+b)
cost = (-1/m)*np.sum(Y*np.log(A) + (1-Y)*np.log(1-A))
# 反向传播
dw = (1/m)* np.dot(X,(A-Y).T)
db = (1/m)* np.sum(A-Y)
assert(dw.shape==W.shape)
assert(db.dtype == float)
cost = np.squeeze(cost)
# 存储dw db
grads = {
"dw":dw,
"db":db
}
return grads, cost
# ### 这里附上课程中求导计算过程中的部分截图,便于理解
# 向量化之前:
# ![image.png](attachment:image.png)
# 向量化后:
# ![image.png](attachment:image.png)
# In[209]:
# 测试 two_way_propogate
# 随便初始化一些参数
W ,b ,X, Y = np.array([ [2],[2] ]),2,np.array([ [1,0],[0,0] ]),np.array([0,0])
grabs, cost = two_way_propogate(W,b,X,Y)
print("dw = " + str(grabs["dw"]))
print("db = " + str(grabs["db"]))
print("cost = " + str(cost))
# 迭代一次 看效果如何
learning_rate = 0.1
W = W -learning_rate*grabs["dw"]
b= b- learning_rate * grabs["db"]
grads, cost = two_way_propogate(W,b,X,Y)
print("迭代一次后")
print("dw = " + str(grabs["dw"]))
print("db = " + str(grabs["db"]))
print("cost = " + str(cost))
# In[210]:
def optimize( W , b , X , Y , num_iteration , learning_rate , print_cost=False , print_frequency = 100):
# 该函数作用 运用梯度下降优化W 和b
# 参数描述放在下面
"""
此函数通过运行梯度下降算法来优化w和b
参数:
w - 权重,大小不等的数组(num_px * num_px * 3,1)
b - 偏差,一个标量
X - 维度为(num_px * num_px * 3,训练数据的数量)的数组。
Y - 真正的“标签”矢量(如果非猫则为0,如果是猫则为1),矩阵维度为(1,训练数据的数量)
num_iterations - 优化循环的迭代次数
learning_rate - 梯度下降更新规则的学习率
print_cost - 每100步打印一次损失值
返回:
params - 包含权重w和偏差b的字典
grads - 包含权重和偏差相对于成本函数的梯度的字典
成本 - 优化期间计算的所有成本列表,将用于绘制学习曲线。
提示:
我们需要写下两个步骤并遍历它们:
1)计算当前参数的成本和梯度,使用propagate()。
2)使用w和b的梯度下降法则更新参数。
"""
costs =[]
for i in range(num_iteration):
grads , cost = two_way_propogate( W , b, X , Y )
dw = grads["dw"]
db = grads["db"]
W = W- dw * learning_rate
b = b- db * learning_rate
if(i%print_frequency==0):
costs.append(cost)
if(print_cost and i%print_frequency==0):
print("迭代次数为:%i , 误差值为: %f" % (i,cost))
params = {
"W": W,
"b": b
}
grads = {
"dw": dw,
"db": db
}
return (params , grabs , costs)
#
w - 权重矩阵
#
b - 偏差值 一个数
#
X - 维度为 训练集
#
Y - 训练标签集 真正的“标签”矢量(如果非猫则为0,如果是猫则为1),矩阵维度为(1,训练数据的数量)
#
num_iterations - 优化循环的迭代次数
#
learning_rate - 梯度下降更新规则的学习率
#
print_cost - 是否打印cost
#
print_frequency - 打印频率,迭代多少次打印一次cost
# In[211]:
# 测试 optimize
W, b, X, Y = np.array([[2], [3]]), 3, np.array([[0,1], [2,2]]), np.array([[0, 2]])
params , grads , costs = optimize(W , b , X , Y , num_iteration=100 , learning_rate = 0.009)
print ("w = " + str(params["W"]))
print ("b = " + str(params["b"]))
print ("dw = " + str(grads["dw"]))
print ("db = " + str(grads["db"]))
# In[212]:
def predict(W , b , X):
# 根据参数 预测结果
m = X.shape[1]
# 这里参数的维度假设是符合的 不需要判断
A = sigmoid( np.dot(W.T,X )+b)
assert(A.shape==(1,m))
for i in range(A.shape[1]):
A[0,i] = 1 if A[0,i] >0.5 else 0
# 默认0.5为临界值
assert(A.shape==(1,m) )
return A
# 这里的0.5 可以修改为一个指定的临界值 或设置为参数
# In[213]:
# 测试predict
w, b, X, Y = np.array([[1], [2]]), 2, np.array([[1,2], [3,4]]), np.array([[1, 0]])
print("predictions = " + str(predict(w, b, X)))
# In[214]:
def model ( X_train , Y_train , X_test , Y_test , num_iterations = 2000 , learning_rate = 0.5, print_cost = False , print_frequency = 100):
W , b = initialize_zeros_vector(X_train.shape[0])
parameters , grabs , costs = optimize(W , b , X_train , Y_train , num_iterations , learning_rate , print_cost, print_frequency)
W , b = parameters["W"] , parameters["b"]
# 预测测试集
Y_prediction_tset = predict(W , b , X_test)
Y_prediction_train = predict(W , b , X_train)
# 打印准确性
print("训练集准确性: " , format(100 - np.mean(np.abs(Y_prediction_train - Y_train)) * 100 ) , "%")
print("测试集准确性: " , format(100 - np.mean(np.abs(Y_prediction_tset - Y_test)) * 100 ) , "%")
d = {
"costs" : costs,
"Y_prediction_test" : Y_prediction_tset,
"Y_prediction_train" : Y_prediction_train,
"W" : W,
"b" : b,
"learning_rate" : learning_rate,
"num_iterations" : num_iterations
}
return d
# In[215]:
# 测数model
d = model(train_set_x,train_set_y,test_set_x,test_set_y,num_iterations=2000,learning_rate=0.005,print_cost=True)
# In[216]:
costs = np.squeeze( d["costs"] )
plt.plot(costs)
plt.ylabel("cost")
plt.xlabel("iterations(per 100)")
plt.title("Learning rate= "+ str(d["learning_rate"]))
plt.show
# In[217]:
# 测试不同的学习率
learning_rates = [0.01,0.001,0.0001]
models = {}
for i in learning_rates:
print("learning_rate is: " + str(i))
models[str(i)]=model(train_set_x,train_set_y,test_set_x,test_set_y,num_iterations=2000,learning_rate=i,print_cost=False)
print("\n")
for i in learning_rates:
plt.plot(np.squeeze(models[str(i)]["costs"] ), label= str(models[str(i)]["learning_rate"]))
plt.ylabel('cost')
plt.xlabel('iterations')
legend = plt.legend(loc='upper center' , shadow=True )
frame = legend.get_frame()
frame.set_facecolor('0.90')
plt.show
# In[218]:
def test_my_image( my_image):
image_name = my_image
fname = "images/" + my_image # 在当前目录下有一个images文件夹 用于存放我们的图片
image = np.array(imageio.imread(fname))
my_image = Image.fromarray(image).resize(( im_size , im_size ))
my_image = np.array(my_image).reshape((1, im_size * im_size * 3)).T
my_predicted_image = predict(d["W"], d["b"], my_image)
plt.imshow(image)
print("image name is: "+image_name)
print( "y = " + str(np.squeeze(my_predicted_image)) + ", your algorithm predicts a \"" + classes[int(np.squeeze(my_predicted_image)),].decode("utf-8") + "\" picture.")
# 下面我们自己整一张图片试一试
# In[220]:
my_image1 = "my_image1.jpg"
# my_image2 = "my_image1.jpg"
# my_image3 = "my_image2.jpg"
# my_image4 = "my_image3.jpg"
# my_image5 = "my_image4.jpg"
test_my_image(my_image1)
# test_my_image(my_image2)
# test_my_image(my_image3)
# test_my_image(my_image4)
# test_my_image(my_image5)
# #### 声明: 本人参考了[Kulbear](https://github.com/Kulbear/deep-learning-coursera) 的github上的文章 ,加以自己理解,编写了本篇内容。尽力让人轻松理解课程内容及作业