RNN处理不定长输入

RNN处理不定长输入

RNN模型：

class RNN(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(RNN, self).__init__()

        self.hidden_size = hidden_size

        self.u = nn.Linear(input_size, hidden_size)
        self.w = nn.Linear(hidden_size, hidden_size)
        self.v = nn.Linear(hidden_size, output_size)

        self.tanh = nn.Tanh()
        self.softmax = nn.LogSoftmax(dim=1)

    def forward(self, inputs, hidden):

        u_x = self.u(inputs)

        hidden = self.w(hidden)
        hidden = self.tanh(hidden + u_x)

        output = self.softmax(self.v(hidden))

        return output, hidden

    def initHidden(self):
        return torch.zeros(1, self.hidden_size)

rnn = RNN(n_input, n_hidden, n_categories)
rnn = RNN(57, 128, 18)

RNN训练过程：

    for i in range(line_tensor.size()[0]):
        output, hidden = rnn(line_tensor[i], hidden)

计算loss后的反向传播：

    loss = criterion(output, category_tensor)
    loss.backward()

梯度下降：

    for p in rnn.parameters():
        p.data.add_(-learning_rate, p.grad.data)

ot：目标分类数，在这里应该是18和下面的labels对应上
这里的xt用的是独热向量，表示58种输入
hidden:

line_tensor:

写个案例：

从csv文件中进行数据生成：

import pandas as pd
import torch
import math
import random
# ============================ step 1/5 数据 ================================
data = pd.read_csv("./data/traffic_28_classifier1.csv")
# data.dropna(axis=0,how='any') #drop all rows that have any NaN values

serises_num=3
serises_len=15
label_tag=13

def arrsToTensor(randomSample):
    tensor = torch.zeros(serises_num, 1, serises_len)  # serises_num是一个样本的序列长度  serises_len是每一个序列的长度  1是pytorch的格式需要
    for i in range(len(randomSample)):
        tensor[i][0] = torch.tensor(randomSample.iloc[i].values)
    return tensor



def randomTrainingExample():
    # 1. 样本数sampleCounts
    # 2. 随机取[1,sampleCounts]中的值randomSample
    # 3. 获取样本数据，data.iloc[randomSampleNum:randomSampleNum+3]
    sampleCounts = data.shape[0]-2
    randomSampleNum = random.randint(1,sampleCounts-2) # 最后一个样本不取，因为没有标签
    #print(randomSampleNum)
    #length_data=data.shape[0]
    #test=data.iloc[0]
    randomSample=data.iloc[randomSampleNum:randomSampleNum+3]
    arrs_tensor = arrsToTensor(randomSample)  # arrs_data
    label_data=data.iloc[randomSampleNum + 3]
    randomSampleLabel=data.iloc[randomSampleNum+3][label_tag] # label
    #print(randomSampleLabel)
    category_tensor = torch.tensor([randomSampleLabel], dtype=torch.long)

    return category_tensor, arrs_tensor

输入为1*15的3个连续向量
输出为0，1的二分类任务

main.py

from model import RNN
from data_loader import randomTrainingExample
import torch.nn as nn
import torch
# ============================ step 0/5 参数设置 ============================
device = torch.device("cpu")
learning_rate = 0.005
n_iters = 200000
n_hidden = 128
n_input = 15
n_categories = 2
# ============================ step 1/5 数据 ================================

# from data_loader import randomTrainingExample

# ============================ step 2/5 模型 ================================


# rnn = RNN(n_input, n_hidden, n_categories)
rnn = RNN(n_input, n_hidden, n_categories)
rnn.to(device)




# ============================ step 3/5 损失函数 ============================

criterion = nn.NLLLoss()

# ============================ step 4/5 优化器 ==============================

# for p in rnn.parameters():  # 优化器
#     p.data.add_(-learning_rate, p.grad.data)


# ============================ step 5/5 训练 ================================

def train(category_tensor, arrs_tensor):
    hidden = rnn.initHidden()

    rnn.zero_grad()

    arrs_tensor = arrs_tensor.to(device)
    hidden = hidden.to(device)
    category_tensor = category_tensor.to(device)

    for i in range(arrs_tensor.size()[0]):
        output, hidden = rnn(arrs_tensor[i], hidden)

    loss = criterion(output, category_tensor)
    loss.backward()

    # Add parameters' gradients to their values, multiplied by learning rate
    for p in rnn.parameters():  # 优化器
        p.data.add_(-learning_rate, p.grad.data)

    return output, loss.item()

for iter in range(1, n_iters + 1):

    category_tensor, arrs_tensor = randomTrainingExample()
    # training
    output, loss = train(category_tensor, arrs_tensor)
    print(loss)


# =========================== inference ====================================

model.py

import torch.nn as nn
import torch

class RNN(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(RNN, self).__init__()

        self.hidden_size = hidden_size

        self.u = nn.Linear(input_size, hidden_size)
        self.w = nn.Linear(hidden_size, hidden_size)
        self.v = nn.Linear(hidden_size, output_size)

        self.tanh = nn.Tanh()
        self.softmax = nn.LogSoftmax(dim=1)

    def forward(self, inputs, hidden):

        u_x = self.u(inputs)

        hidden = self.w(hidden)
        hidden = self.tanh(hidden + u_x)

        output = self.softmax(self.v(hidden))

        return output, hidden

    def initHidden(self):
        return torch.zeros(1, self.hidden_size)

data_loader.py

import pandas as pd
import torch
import math
import random
# ============================ step 1/5 数据 ================================
data = pd.read_csv("./data/traffic_28_classifier1.csv")
# data.dropna(axis=0,how='any') #drop all rows that have any NaN values

serises_num=3
serises_len=15
label_tag=13

def arrsToTensor(randomSample):
    tensor = torch.zeros(serises_num, 1, serises_len)  # serises_num是一个样本的序列长度  serises_len是每一个序列的长度  1是pytorch的格式需要
    for i in range(len(randomSample)):
        tensor[i][0] = torch.tensor(randomSample.iloc[i].values)
    return tensor



def randomTrainingExample():
    # 1. 对所有的行数除以3，向下取整得出样本数sampleCounts
    # 2. 随机取[1,sampleCounts]中的值randomSample
    # 3. 获取样本数据，randomSample*3, data[randomSample*3-3,randomSample*3]
    sampleCounts = data.shape[0]-2
    randomSampleNum = random.randint(1,sampleCounts-2) # 最后一个样本不取，因为没有标签
    #print(randomSampleNum)
    #length_data=data.shape[0]
    #test=data.iloc[0]
    randomSample=data.iloc[randomSampleNum:randomSampleNum+3]
    arrs_tensor = arrsToTensor(randomSample)  # arrs_data
    label_data=data.iloc[randomSampleNum + 3]
    randomSampleLabel=data.iloc[randomSampleNum+3][label_tag] # label
    #print(randomSampleLabel)
    category_tensor = torch.tensor([randomSampleLabel], dtype=torch.long)

    return category_tensor, arrs_tensor


# def traffic_data_loader():
#     # 每取三行，对第四行的traffic进行判断，如果第四行的traffic是1，就是正样例，如果traffic=0,就是负样例
#
#     positive = pd.DataFrame()
#     negective = pd.DataFrame()
#
#     for i in range(3,data.shape[0]):
#         if data.iloc[i][13]== 1:
#             positive=positive.append(data.iloc[i-3:i])
#             #print(data.iloc[i-3:i])
#         elif data.iloc[i][13]== 0:
#             negective=negective.append(data.iloc[i-3:i])
#             #print(data.iloc[i-3:i])
#
#     negective.to_csv("negective.csv")
#     positive.to_csv("positive.csv")
#
#     data_single=data.iloc[1]
#     data_single=data_single.values
#     data_single
#     t=torch.tensor(data_single)
#
#
#     tensor = torch.zeros(3, 1, 15) # 3是一个样本的序列长度  15是每一个序列的长度  1是pytorch的格式需要
#     tensor[0][1]= t;





    # return  tensor

if __name__ == "__main__":

   category_tensor, arrs_tensor=randomTrainingExample()
   print(category_tensor, arrs_tensor)