3.神经网络
深度学习入门
本文的文件和代码链接:github地址
1.激活函数
- sigmoid
h ( x ) = 1 1 + e − x h(x) = \frac{1}{1 + e^{-x}} h(x)=1+e−x1
def sigmod(x):
return 1 / (1 + np.exp(-1 * x))
- ReLU
h ( x ) = { x : x > 0 0 : x ≤ 0 h(x) = \left\{ \begin{array}{lr} x & : x > 0\\ 0 & : x \le 0 \end{array} \right. h(x)={x0:x>0:x≤0
- softmax 函数(常用来分类)
y k = e a k ∑ i = 1 n e a i y_k = \frac{e^{a_k}}{\sum_{i=1}^n e^{a_i}} yk=∑i=1neaieak
需要关注:softmax需要进行指数运算,因此容易溢出
解决方法:
y k = e a k ∑ i = 1 n e a i = C ∗ e a k C ∗ ∑ i = 1 n e a i = e x p ( a k + l o g C ) ∗ ∑ i = 1 n e x p ( a i + l o g C ) = e x p ( a k + C ′ ) ∗ ∑ i = 1 n e x p ( a i + C ′ ) y_k = \frac{e^{a_k}}{\sum_{i=1}^n e^{a_i}} = \frac{C *e^{a_k}}{C * \sum_{i= 1}^n e^{a_i}} = \frac{exp(a_k + logC)}{* \sum_{i= 1}^n exp(a_i + logC)} = \frac{exp(a_k + C')}{* \sum_{i= 1}^n exp(a_i + C')} yk=∑i=1neaieak=C∗∑i=1neaiC∗eak=∗∑i=1nexp(ai+logC)exp(ak+logC)=∗∑i=1nexp(ai+C′)exp(ak+C′)
即在进行softmax指数运算的时候,加上或减去某个数,结果不变,因此可以减去输入信号中的最大值
softmax代码实现:
def softmax(a):
c = np.max(a)
return np.exp(a - c) / np.sum(np.exp(a - c)) # 利用了数组的广播机制
2. 使用mnist数据集进行推理
数据集导入
import sys, os
import numpy as np
# 为了导入父目录中的文件, 即将父目录加入到 sys.path(Python的搜索模块)的路径集中
sys.path.append(os.pardir)
# 其中 dataset.mnist 为dataset文件夹下的python文件,用来进行数据集的预处理
from dataset.mnist import load_mnist
# 下载数据集
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True, flatten=True, one_hot_label=True)
显示mnist图像
img = x_train[0]
label = t_train[0]
# 将图像形状转为(1, 28, 28)
img = img.reshape(28, 28)
# 使用 matplotlib.pyplot 进行查看
import matplotlib.pyplot as plt
plt.imshow(img)
显示结果:
前向推理过程
函数定义:
import pickle
# 因为只是进行测试,所以只需要获取测试集的数据
def get_data():
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True, flatten=True, one_hot_label=True)
return x_test, t_test
# 初始化网络,从文件中读取之前保存好的权重(因为此时还没有学习如何进行训练,只是进行推理,因此使用给定的参数进行推理)
def init_network(file_path):
with open(file_path, 'rb') as f:
network = pickle.load(f)
return network
# 进行推理预测
def predict(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1
z1 = sigmod(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmod(a2)
a3 = np.dot(z2, W3) + b3
z3 = softmax(a3)
return z3
进行推理:
# 进行推理
x, t = get_data()
network = init_network("sample_weight.pkl")
# cnt 统计预测正确的个数
cnt = 0
# 遍历每一个样本
for i in range(x.shape[0]):
y = predict(network, x[i])
h = np.argmax(y) # 获取y中最大值的索引
if h == np.argmax(t[i]):
cnt += 1
# cnt 最终输出为 9352
3. 批处理
之前预测的过程中一次处理一个样本,现在考虑一次处理多个样本的情况,即批处理。
一次打包输入多张图片(一张图片是一个样本,多张图片就是多个样本),这种打包式的输入就被称为批。
# 进行推理
x, t = get_data()
network = init_network("sample_weight.pkl")
# batch_size 定义一批处理的样本数
batch_size = 100
# cnt 统计预测正确的个数
cnt = 0
# 遍历每一个样本
for i in range(0, x.shape[0], batch_size):
y = predict(network, x[i:i+batch_size])
h = np.argmax(y, axis = 1) # 按照列, 获取y中每一行中最大值的索引(行不变,在列上计算, 因此axis=1)
cnt += np.sum(h == np.argmax(t[i:i+batch_size], axis = 1))
# cnt仍然为 9352
4. 补充说明
dataset目录下 mnist.py 文件内容:
# coding: utf-8
try:
import urllib.request
except ImportError:
raise ImportError('You should use Python 3.x')
import os.path
import gzip
import pickle
import os
import numpy as np
url_base = 'http://yann.lecun.com/exdb/mnist/'
key_file = {
'train_img':'train-images-idx3-ubyte.gz',
'train_label':'train-labels-idx1-ubyte.gz',
'test_img':'t10k-images-idx3-ubyte.gz',
'test_label':'t10k-labels-idx1-ubyte.gz'
}
dataset_dir = os.path.dirname(os.path.abspath(__file__))
save_file = dataset_dir + "/mnist.pkl"
train_num = 60000
test_num = 10000
img_dim = (1, 28, 28)
img_size = 784
def _download(file_name):
file_path = dataset_dir + "/" + file_name
if os.path.exists(file_path):
return
print("Downloading " + file_name + " ... ")
urllib.request.urlretrieve(url_base + file_name, file_path)
print("Done")
def download_mnist():
for v in key_file.values():
# 其中 v 是 key_file 中的值, 不是key
_download(v) # 下载后文件名为 /train-images-idx3-ubyte.gz 等
def _load_label(file_name):
file_path = dataset_dir + "/" + file_name
print("Converting " + file_name + " to NumPy Array ...")
with gzip.open(file_path, 'rb') as f:
labels = np.frombuffer(f.read(), np.uint8, offset=8)
print("Done")
return labels
def _load_img(file_name):
# 此时 file_path 为 /train-images-idx3-ubyte.gz等
file_path = dataset_dir + "/" + file_name
print("Converting " + file_name + " to NumPy Array ...")
with gzip.open(file_path, 'rb') as f:
# np.frombuffer 将缓冲区解释为一维数组, 即将 /train-images-idx3-ubyte.gz 解释为一维数组
data = np.frombuffer(f.read(), np.uint8, offset=16)
data = data.reshape(-1, img_size)
print("Done")
return data
# 将下载后的对象转为 numpy
def _convert_numpy():
dataset = {}
dataset['train_img'] = _load_img(key_file['train_img'])
dataset['train_label'] = _load_label(key_file['train_label'])
dataset['test_img'] = _load_img(key_file['test_img'])
dataset['test_label'] = _load_label(key_file['test_label'])
return dataset
def init_mnist():
download_mnist()
dataset = _convert_numpy()
print("Creating pickle file ...")
with open(save_file, 'wb') as f:
# 序列化操作,将对象dataset保存到 f 文件中,其中 f为 dataset_dir + "/mnist.pkl"
pickle.dump(dataset, f, -1)
print("Done!")
def _change_one_hot_label(X):
T = np.zeros((X.size, 10))
for idx, row in enumerate(T):
row[X[idx]] = 1
return T
def load_mnist(normalize=True, flatten=True, one_hot_label=False):
"""读入MNIST数据集
Parameters
----------
normalize : 将图像的像素值正规化为0.0~1.0
one_hot_label :
one_hot_label为True的情况下,标签作为one-hot数组返回
one-hot数组是指[0,0,1,0,0,0,0,0,0,0]这样的数组
flatten : 是否将图像展开为一维数组
Returns
-------
(训练图像, 训练标签), (测试图像, 测试标签)
"""
if not os.path.exists(save_file):
init_mnist()
with open(save_file, 'rb') as f:
dataset = pickle.load(f)
if normalize:
for key in ('train_img', 'test_img'):
dataset[key] = dataset[key].astype(np.float32)
dataset[key] /= 255.0
if one_hot_label:
dataset['train_label'] = _change_one_hot_label(dataset['train_label'])
dataset['test_label'] = _change_one_hot_label(dataset['test_label'])
if not flatten:
for key in ('train_img', 'test_img'):
dataset[key] = dataset[key].reshape(-1, 1, 28, 28)
return (dataset['train_img'], dataset['train_label']), (dataset['test_img'], dataset['test_label'])
if __name__ == '__main__':
init_mnist()