为了账号安全,请及时绑定邮箱和手机立即绑定

深度学习之美——CNN,RNN

标签:
深度学习

一.神经网络初识

1.神经元的二分类
图片描述
2.神经元的多分类
图片描述

此时的W不再是一维向量,而是二维矩阵

分为k类

图片描述

图片描述

代码实例:
Tensorflow实现多层神经元
数据集:
图片描述

训练集:data_batch
图片描述

图片描述

图片描述

图片描述

Data的大小 : 10000 X 3072
说明有10000张图片
为什么是3072: 32X32(像素) X 3(通道) = 3072

Labels:一共有10类图片
图片描述

展开并打印Data数据中的一张图片
图片描述

图片描述

不进行维度转换就会报错:

图片描述

读数据:

import os 
import numpy as np  
import tensorflow as tf 
import pickle 
import matplotlib.pyplot as plt 
from  matplotlib.pyplot import imshow  

CIFAR_DIR = "./cifar-10-batches-py"
	

#read_data from file 
def load_data(filename):
	with open(filename,'rb') as f:
		data = pickle.load(f, encoding='bytes')  
		return data[b'data'],data[b'labels']

#数据预处理
class CifarData:
	def __init__(self,filenames,need_shuffle):
		all_data = []
		all_labels = []
		for filename in filenames:
			data,labels = load_data(filename)
			#把数据绑定到一起
			for item,label in zip(data,labels):
				if label in [0,1]:
					all_data.append(item)
					all_labels.append(label)
		self._data = np.vstack(all_data) #纵向合并
		self._labels = np.hstack(all_labels) #横向合并
		print (self._data.shape)
		print (self._labels.shape)
		self._num_examples = self._data.shape[0]
		self._need_shuffle = need_shuffle
		#_indicator: 数据集起始位置
		self._indicator = 0 

		if self._need_shuffle:
			self._shuffle_data()

	def _shuffle_data(self):
		#打乱
		p = np.random.permutation(self._num_examples)
		self._data = self._data[p]
		self._labels = self._labels[p] 

	def next_batch(self,batch_size):
		#更新起始位置
		end_indicator = self._indicator + batch_size
		if end_indicator > self._num_examples:
			if self._need_shuffle:
				self._shuffle_data()
				#重置
				self._indicator = 0 
				end_indicator   = batch_size
			else:
				raise Exception("Have no more examples") 

		#保险起见,再判断一遍
		if end_indicator > self._num_examples:
			raise Exception("batch size is larger than all examples")
		batch_data = self._data[self._indicator : end_indicator]
		batch_labels = self._labels[self._indicator : end_indicator]
		self._indicator = end_indicator
		return batch_data,batch_labels 

train_filenames = [os.path.join(CIFAR_DIR,'data_batch_%d' % i) for i in range(1,6)]
test_filenames =  [os.path.join(CIFAR_DIR,'test_batch')]
train_data  = CifarData(train_filenames,True)
batch_data,batch_labels = train_data.next_batch(10)
print(batch_data)
print(batch_labels)

运行结果:
图片描述

实现神经元——二分类:

import os 
import numpy as np  
import tensorflow as tf 
import pickle 
import matplotlib.pyplot as plt 
from  matplotlib.pyplot import imshow  

CIFAR_DIR = "./cifar-10-batches-py"
	

#read_data from file 
def load_data(filename):
	with open(filename,'rb') as f:
		data = pickle.load(f, encoding='bytes')  
		return data[b'data'],data[b'labels']

#数据预处理
class CifarData:
	def __init__(self,filenames,need_shuffle):
		all_data = []
		all_labels = []
		for filename in filenames:
			data,labels = load_data(filename)
			#把数据绑定到一起
			for item,label in zip(data,labels):
				if label in [0,1]:
					all_data.append(item)
					all_labels.append(label)
		self._data = np.vstack(all_data) #纵向合并
		#特征值缩放到0,1之间
		self._data = self._data / 127.5 - 1 
		self._labels = np.hstack(all_labels) #横向合并
		print (self._data.shape)
		print (self._labels.shape)
		self._num_examples = self._data.shape[0]
		self._need_shuffle = need_shuffle
		#_indicator: 数据集起始位置
		self._indicator = 0 

		if self._need_shuffle:
			self._shuffle_data()

	def _shuffle_data(self):
		#打乱
		p = np.random.permutation(self._num_examples)
		self._data = self._data[p]
		self._labels = self._labels[p] 

	def next_batch(self,batch_size):
		#更新起始位置
		end_indicator = self._indicator + batch_size
		if end_indicator > self._num_examples:
			if self._need_shuffle:
				self._shuffle_data()
				#重置
				self._indicator = 0 
				end_indicator   = batch_size
			else:
				raise Exception("Have no more examples") 

		#保险起见,再判断一遍
		if end_indicator > self._num_examples:
			raise Exception("batch size is larger than all examples")
		batch_data = self._data[self._indicator : end_indicator]
		batch_labels = self._labels[self._indicator : end_indicator]
		self._indicator = end_indicator
		return batch_data,batch_labels 

train_filenames = [os.path.join(CIFAR_DIR,'data_batch_%d' % i) for i in range(1,6)]
test_filenames =  [os.path.join(CIFAR_DIR,'test_batch')]
train_data  = CifarData(train_filenames,True)
batch_data,batch_labels = train_data.next_batch(10)



#搭建tensorflow计算图  
x = tf.placeholder(tf.float32,[None,3072])
y = tf.placeholder(tf.int64,[None])

#权重
w = tf.get_variable('w',[x.get_shape()[-1],1], initializer = tf.random_normal_initializer(0,1))
b = tf.get_variable('b',[1],initializer = tf.constant_initializer(0.0))
# [None , 3072] * [3072 , 1] = [None,1]
y_ = tf.matmul(x,w) + b  

#用sigmoid 把y_转换成概率
#p_y_1,y_ : [None,1]
p_y_1 = tf.nn.sigmoid(y_) 

#y: [None] --> [None,1]
y_reshaped = tf.reshape(y,(-1,1))
#y:int64 --> float32
y_reshaped_float = tf.cast(y_reshaped,tf.float32)
#tf对数据类型敏感,不转换会出问题

loss = tf.reduce_mean(tf.square(y_reshaped_float - p_y_1))

predict = p_y_1 > 0.5 
#>0.5 : True,<0.5: False
correct_prediction = tf.equal(tf.cast(predict,tf.int64),y_reshaped)
#求平均
accuracy =  tf.reduce_mean(tf.cast(correct_prediction,tf.float64))

#梯度下降
with tf.name_scope("train_op"):
	train_op = tf.train.AdamOptimizer(1e-3).minimize(loss) 

init = tf.global_variables_initializer()


batch_size = 20 
train_steps = 10000
test_steps = 100

with tf.Session() as sess:
	sess.run(init)
	for i in range(train_steps):
		batch_data,batch_labels = train_data.next_batch(batch_size) 
		loss_val,accu_val, _ = sess.run([loss,accuracy,train_op] , feed_dict = {x:batch_data , y:batch_labels})
		if i % 200 == 0:
			print ('[Train] Step: %d, loss: %4.5f, acc: %4.5f' % (i, loss_val,accu_val))
		if i % 1000 == 0:
			test_data = CifarData(test_filenames,False) 
			all_test_acc_val = []
			for j in range(test_steps):
				test_batch_data,test_batch_labels = test_data.next_batch(batch_size)
				test_acc_val = sess.run( 
					   [accuracy], feed_dict = {x:test_batch_data, y:test_batch_labels}
					)
				all_test_acc_val.append(test_acc_val)
			test_acc = np.mean(all_test_acc_val)
			print ('[Test] Step: %d, acc: %4.5f' % (i,test_acc))


	

运行结果 :

图片描述

实现神经元——多分类:

import os 
import numpy as np  
import tensorflow as tf 
import pickle 
import matplotlib.pyplot as plt 
from  matplotlib.pyplot import imshow  

CIFAR_DIR = "./cifar-10-batches-py"
	

#read_data from file 
def load_data(filename):
	with open(filename,'rb') as f:
		data = pickle.load(f, encoding='bytes')  
		return data[b'data'],data[b'labels']

#数据预处理
class CifarData:
	def __init__(self,filenames,need_shuffle):
		all_data = []
		all_labels = []
		for filename in filenames:
			data,labels = load_data(filename)
			#把数据绑定到一起
			all_data.append(data)
			all_labels.append(labels)



		self._data = np.vstack(all_data) #纵向合并
		#特征值缩放到0,1之间
		self._data = self._data / 127.5 - 1 
		self._labels = np.hstack(all_labels) #横向合并
		print (self._data.shape)
		print (self._labels.shape)
		self._num_examples = self._data.shape[0]
		self._need_shuffle = need_shuffle
		#_indicator: 数据集起始位置
		self._indicator = 0 

		if self._need_shuffle:
			self._shuffle_data()

	def _shuffle_data(self):
		#打乱
		p = np.random.permutation(self._num_examples)
		self._data = self._data[p]
		self._labels = self._labels[p] 

	def next_batch(self,batch_size):
		#更新起始位置
		end_indicator = self._indicator + batch_size
		if end_indicator > self._num_examples:
			if self._need_shuffle:
				self._shuffle_data()
				#重置
				self._indicator = 0 
				end_indicator   = batch_size
			else:
				raise Exception("Have no more examples") 

		#保险起见,再判断一遍
		if end_indicator > self._num_examples:
			raise Exception("batch size is larger than all examples")
		batch_data = self._data[self._indicator : end_indicator]
		batch_labels = self._labels[self._indicator : end_indicator]
		self._indicator = end_indicator
		return batch_data,batch_labels 

train_filenames = [os.path.join(CIFAR_DIR,'data_batch_%d' % i) for i in range(1,6)]
test_filenames =  [os.path.join(CIFAR_DIR,'test_batch')]
train_data  = CifarData(train_filenames,True)
batch_data,batch_labels = train_data.next_batch(10)



#搭建tensorflow计算图  
x = tf.placeholder(tf.float32,[None,3072])
y = tf.placeholder(tf.int64,[None])

#权重
w = tf.get_variable('w',[x.get_shape()[-1],10], initializer = tf.random_normal_initializer(0,1))
b = tf.get_variable('b',[10],initializer = tf.constant_initializer(0.0))
# [None , 3072] * [3072 , 10] = [None,10]
y_ = tf.matmul(x,w) + b  

#softmax而不是sigmoid表示概率 
#softmax = tf.exp(logits) / tf.reduce_sum(tf.exp(logits) ,axis)

"""
#Method1,使用平方差损失函数
p_y_1 = tf.nn.softmax(y_) 
#结合softmax的特点,为了计算损失函数,要对y进行one-hot编码
y_one_hot = tf.one_hot(y,10,dtype = tf.float32) 

loss = tf.reduce_mean(tf.square(y_one_hot - p_y_1))
"""

#Method2,使用交叉熵损失函数 
loss = tf.losses.sparse_softmax_cross_entropy(labels = y,logits = y_)
#该函数完成三个步骤
#1.y_ -> softmax 
#2.y  -> one_hot 
#3.y*logy_


#求最大值
predict = tf.argmax(y_,1)
correct_prediction = tf.equal(predict,y)
#求平均
accuracy =  tf.reduce_mean(tf.cast(correct_prediction,tf.float64))

#梯度下降
with tf.name_scope("train_op"):
	train_op = tf.train.AdamOptimizer(1e-3).minimize(loss) 

init = tf.global_variables_initializer()


batch_size = 20 
train_steps = 10000
test_steps = 100

with tf.Session() as sess:
	sess.run(init)
	for i in range(train_steps):
		batch_data,batch_labels = train_data.next_batch(batch_size) 
		loss_val,accu_val, _ = sess.run([loss,accuracy,train_op] , feed_dict = {x:batch_data , y:batch_labels})
		if i % 200 == 0:
			print ('[Train] Step: %d, loss: %4.5f, acc: %4.5f' % (i, loss_val,accu_val))
		if i % 1000 == 0:
			test_data = CifarData(test_filenames,False) 
			all_test_acc_val = []
			for j in range(test_steps):
				test_batch_data,test_batch_labels = test_data.next_batch(batch_size)
				test_acc_val = sess.run( 
					   [accuracy], feed_dict = {x:test_batch_data, y:test_batch_labels}
					)
				all_test_acc_val.append(test_acc_val)
			test_acc = np.mean(all_test_acc_val)
			print ('[Test] Step: %d, acc: %4.5f' % (i,test_acc))


运行结果:
图片描述

一个神经元,求梯度,求一次偏导
多层神经元,相当于复合函数,求偏导时,用的是多元微分学中的链式法则

图片描述

图片描述
反向传播存在的问题:
图片描述

优化方法:
图片描述
优化后还存在振荡和局部最优解问题:
解决办法: 和一般梯度下降对比

图片描述
好处:
图片描述

二,卷积神经网络

普通神经网络遇到的问题:
图片描述

图片描述

图片描述
卷积的计算过程:

图片描述

图片描述

之前的卷积都是单通道一个卷积核的,下面看看多通道多卷积核的卷积运算:
三通道一个卷积核:相当于三层卷积核

图片描述

三通道六个卷积核:卷积核与卷积核之间,不共享数据

图片描述

一个卷积核对应图像的一种特征

做一个计算题:

图片描述

尺寸计算:
图片描述

参数数目计算:
图片描述

两种池化方式:
图片描述

图片描述

完整代码:

import os 
import numpy as np  
import tensorflow as tf 
import pickle 
import matplotlib.pyplot as plt 
from  matplotlib.pyplot import imshow  

CIFAR_DIR = "./cifar-10-batches-py"
	

#read_data from file 
def load_data(filename):
	with open(filename,'rb') as f:
		data = pickle.load(f, encoding='bytes')  
		return data[b'data'],data[b'labels']

#数据预处理
class CifarData:
	def __init__(self,filenames,need_shuffle):
		all_data = []
		all_labels = []
		for filename in filenames:
			data,labels = load_data(filename)
			#把数据绑定到一起
			all_data.append(data)
			all_labels.append(labels)



		self._data = np.vstack(all_data) #纵向合并
		#特征值缩放到0,1之间
		self._data = self._data / 127.5 - 1 
		self._labels = np.hstack(all_labels) #横向合并
		print (self._data.shape)
		print (self._labels.shape)
		self._num_examples = self._data.shape[0]
		self._need_shuffle = need_shuffle
		#_indicator: 数据集起始位置
		self._indicator = 0 

		if self._need_shuffle:
			self._shuffle_data()

	def _shuffle_data(self):
		#打乱
		p = np.random.permutation(self._num_examples)
		self._data = self._data[p]
		self._labels = self._labels[p] 

	def next_batch(self,batch_size):
		#更新起始位置
		end_indicator = self._indicator + batch_size
		if end_indicator > self._num_examples:
			if self._need_shuffle:
				self._shuffle_data()
				#重置
				self._indicator = 0 
				end_indicator   = batch_size
			else:
				raise Exception("Have no more examples") 

		#保险起见,再判断一遍
		if end_indicator > self._num_examples:
			raise Exception("batch size is larger than all examples")
		batch_data = self._data[self._indicator : end_indicator]
		batch_labels = self._labels[self._indicator : end_indicator]
		self._indicator = end_indicator
		return batch_data,batch_labels 

train_filenames = [os.path.join(CIFAR_DIR,'data_batch_%d' % i) for i in range(1,6)]
test_filenames =  [os.path.join(CIFAR_DIR,'test_batch')]
train_data  = CifarData(train_filenames,True)
batch_data,batch_labels = train_data.next_batch(10)



#搭建tensorflow计算图  
x = tf.placeholder(tf.float32,[None,3072])
y = tf.placeholder(tf.int64,[None])

#由于要输入三通道的图像,对x进行维度转换
x_image = tf.reshape(x,[-1,3,32,32])
x_image = tf.transpose(x_image,perm = [0,2,3,1])

#CNN
conv1 = tf.layers.conv2d(x_image,
						32,       #卷积核通道数
						(3,3),     #卷积核大小
						padding = 'same', #same表示输出的神经元大小不变
						activation = tf.nn.relu,
						name = 'conv1'
	                    )

pooling1 = tf.layers.max_pooling2d(conv1,
								  (2,2),#核的大小
								  (2,2),#步长
								  name = 'pool1'
	                              )

conv2 = tf.layers.conv2d(pooling1,
						32,       #卷积核通道数
						(3,3),     #卷积核大小
						padding = 'same', #same表示输出的神经元大小不变
						activation = tf.nn.relu,
						name = 'conv2'
	                    )

pooling2 = tf.layers.max_pooling2d(conv2,
								  (2,2),#核的大小
								  (2,2),#步长
								  name = 'pool2'
	                              )

conv3 = tf.layers.conv2d(pooling2,
						32,       #卷积核通道数
						(3,3),     #卷积核大小
						padding = 'same', #same表示输出的神经元大小不变
						activation = tf.nn.relu,
						name = 'conv3'
	                    )

pooling3 = tf.layers.max_pooling2d(conv3,
								  (2,2),#核的大小
								  (2,2),#步长
								  name = 'pool3'
	                              )

#展开
#[None,4*4*32]
flatten = tf.layers.flatten(pooling3) 

#全连接层:变成一维
y_ = tf.layers.dense(flatten,10)


#Method2,使用交叉熵损失函数 
loss = tf.losses.sparse_softmax_cross_entropy(labels = y,logits = y_)


#求最大值
predict = tf.argmax(y_,1)
correct_prediction = tf.equal(predict,y)
#求平均
accuracy =  tf.reduce_mean(tf.cast(correct_prediction,tf.float64))

#梯度下降
with tf.name_scope("train_op"):
	train_op = tf.train.AdamOptimizer(1e-3).minimize(loss) 

init = tf.global_variables_initializer()


batch_size = 20 
train_steps = 10000
test_steps = 100

with tf.Session() as sess:
	sess.run(init)
	for i in range(train_steps):
		batch_data,batch_labels = train_data.next_batch(batch_size) 
		loss_val,accu_val, _ = sess.run([loss,accuracy,train_op] , feed_dict = {x:batch_data , y:batch_labels})
		if i % 200 == 0:
			print ('[Train] Step: %d, loss: %4.5f, acc: %4.5f' % (i, loss_val,accu_val))
		if i % 1000 == 0:
			test_data = CifarData(test_filenames,False) 
			all_test_acc_val = []
			for j in range(test_steps):
				test_batch_data,test_batch_labels = test_data.next_batch(batch_size)
				test_acc_val = sess.run( 
					   [accuracy], feed_dict = {x:test_batch_data, y:test_batch_labels}
					)
				all_test_acc_val.append(test_acc_val)
			test_acc = np.mean(all_test_acc_val)
			print ('[Test] Step: %d, acc: %4.5f' % (i,test_acc))

图片描述

三,循环神经网络

图片描述

图片描述

复合函数展开:
图片描述
反向传播存在的问题:
图片描述
分块儿的梯度计算方式:
图片描述

玩点刺激的,多层神经网络,别头晕!!

图片描述

图片描述

当前很时髦的技术,LSTM:
图片描述

单元解析:

图片描述

图片描述

图片描述

图片描述

从“主语”,到描述“主语”的信息,再到“主语”后面的“动词”,完整的描述流程。

基于LSTM的文本分类模型:

图片描述

图片描述

从词语到句子的HAN算法:

图片描述

词语经过类似:wx + b 的加权,形成句子的编码 ,然后句子经过类似:wx + b 的加权,形成最后的结果。

基于CNN的文本分类模型:
图片描述
图片描述

图片描述

每个词用6维的向量表示
图片描述
RNN和CNN在文本处理上的异同点:
图片描述
RNN+CNN融合使用 :
图片描述
处理流程:在RNN中嵌入卷积运算

图片描述

词向量压缩:为了减少参数,防止过拟合

图片描述

参考教程:
一位大佬的课程,很推荐
https://coding.imooc.com/lesson/259.html#mid=16497

第50篇博客了,从学习算法到现在一直是搬运工,咬咬牙坚持下去,希望以后能有自己的输出,感谢慕课网这个平台。😎

点击查看更多内容
TA 点赞

若觉得本文不错,就分享一下吧!

评论

作者其他优质文章

正在加载中
  • 推荐
  • 评论
  • 收藏
  • 共同学习,写下你的评论
感谢您的支持,我会继续努力的~
扫码打赏,你说多少就多少
赞赏金额会直接到老师账户
支付方式
打开微信扫一扫,即可进行扫码打赏哦
今天注册有机会得

100积分直接送

付费专栏免费学

大额优惠券免费领

立即参与 放弃机会
意见反馈 帮助中心 APP下载
官方微信

举报

0/150
提交
取消