1.数据来源：本例将会使用一个国外的共享单车公开数据集（Capital Bikeshare）来完成我们的任务，数据集下载链接：www.capitalbikeshare.com/ system-data。（或者也在我上传的资源中也有https://download.csdn.net/download/qq_40840797/20325952）

数据介绍：特征因素（例如：年份，季度，星期几，温度，风速等因素）；目标变量（用户数，临时注册用户数，注册用户数）

 2数据处理

import numpy as np
import pandas as pd #读取csv文件的库
from matplotlib import pyplot as plt
import torch
import torch.optim as optim
data_path = 'Bike-Sharing-Dataset/hour.csv'#数据位置
rides = pd.read_csv(data_path)#读取数据文件
rides.head()#看看数据长什么样子

  2.1.原始数据的星期几，天气情况等因素需要编码，这样数据才有意义。

# season=1,2,3,4, weathersi=1,2,3, mnth= 1,2,...,12, hr=0,1, ...,23, weekday=0,1,...,6
# 经过下面的处理后，将会多出若干特征，例如，对于season变量就会有 season_1, season_2, season_3, season_4

dummy_fields = ['season', 'weathersit', 'mnth', 'hr', 'weekday']# 这四种不同的特征,进行编码
for each in dummy_fields:
    #利用pandas对象，我们可以很方便地将一个类型变量属性进行one-hot编码，变成多个属性
    dummies = pd.get_dummies(rides[each], prefix=each, drop_first=False)#prefix是用来生成列名字，drop_first是否把每次生成的第一列去掉
    rides = pd.concat([rides, dummies], axis=1)#按照列方向合并

2.2.把原来的表格里面被编码原始列与不相关列去掉

# 把原有的类型变量对应的特征去掉，将一些不相关的特征去掉
fields_to_drop = ['instant', 'dteday', 'season', 'weathersit', 
                  'weekday', 'atemp', 'mnth', 'workingday', 'hr']#要去掉的列名字
data = rides.drop(fields_to_drop, axis=1)#从文件中去掉
data.head()

2.4.归一化  'cnt', 'temp', 'hum', 'windspeed'这四个列

quant_features = ['cnt', 'temp', 'hum', 'windspeed']#要归一化的四列
#quant_features = ['temp', 'hum', 'windspeed']

# 我们将每一个变量的均值和方差都存储到scaled_features变量中。
scaled_features = {}#空集合
for each in quant_features:
    mean, std = data[each].mean(), data[each].std()#求均值和方差
    scaled_features[each] = [mean, std]#将均值和方差都装进集合
    data.loc[:, each] = (data[each] - mean)/std#求归一化结果并装载进data

 2.5.训练数据与测试数据划分

train_data = data[:-21*24]
test_data = data[-21*24:]#最后21天是测试集

print('训练数据：',len(train_data),'测试数据：',len(test_data))

2.6生成训练样本与训练目标值

target_fields = ['cnt', 'casual', 'registered']#训目标列【用户数，临时用户数，注册用户数】
features, targets = train_data.drop(target_fields, axis=1), train_data[target_fields]
test_features, test_targets = test_data.drop(target_fields, axis=1), test_data[target_fields]

# 将数据从pandas dataframe转换为numpy
X = features.values#取值
Y = targets['cnt'].values#取值
Y = Y.astype(float)#转化格式

Y = np.reshape(Y, [len(Y),1])#【16875，】转变为【16875，1】
losses = []
features.head()

3.两种构建神经网络方式

   3.1.手动编写神经网络

input_size = features.shape[1] #输入层单元个数
hidden_size = 10 #隐含层单元个数
output_size = 1 #输出层单元个数
batch_size = 128 #每隔batch的记录数
weights1 = torch.randn([input_size, hidden_size], dtype = torch.double,  requires_grad = True) #第一到二层权重
biases1 = torch.randn([hidden_size], dtype = torch.double, requires_grad = True) #隐含层偏置
weights2 = torch.randn([hidden_size, output_size], dtype = torch.double, requires_grad = True) #隐含层到输出层权重
def neu(x):
    #计算隐含层输出
    #x为batch_size * input_size的矩阵，weights1为input_size*hidden_size矩阵，
    #biases为hidden_size向量，输出为batch_size * hidden_size矩阵    
    hidden = x.mm(weights1) + biases1.expand(x.size()[0], hidden_size)
    hidden = torch.sigmoid(hidden)
    
    #输入batch_size * hidden_size矩阵，mm上weights2, hidden_size*output_size矩阵，
    #输出batch_size*output_size矩阵
    output = hidden.mm(weights2)
    return output
def cost(x, y):
    # 计算损失函数
    error = torch.mean((x - y)**2)
    return error
def zero_grad():
    # 清空每个参数的梯度信息
    if weights1.grad is not None and biases1.grad is not None and weights2.grad is not None:
        weights1.grad.data.zero_()
        weights2.grad.data.zero_()
        biases1.grad.data.zero_()
def optimizer_step(learning_rate):
    # 梯度下降算法
    weights1.data.add_(- learning_rate * weights1.grad.data)
    weights2.data.add_(- learning_rate * weights2.grad.data)
    biases1.data.add_(- learning_rate * biases1.grad.data)
losses = []
for i in range(1000):
    # 每128个样本点被划分为一个撮，在循环的时候一批一批地读取
    batch_loss = []#每一批次的损失
    # start和end分别是提取一个batch数据的起始和终止下标
    for start in range(0, len(X), batch_size):#每128个点作为一个批次
        end = start + batch_size if start + batch_size < len(X) else len(X)#取截止点位
        xx = torch.tensor(X[start:end], dtype = torch.double, requires_grad = True)#生成特征批次
        yy = torch.tensor(Y[start:end], dtype = torch.double, requires_grad = True)#生成目标批次
        predict = neu(xx)#预测
        loss = cost(predict, yy)#损失
        zero_grad()#梯度清零
        loss.backward()#损失反向传播
        optimizer_step(0.01)#学习率设置
        batch_loss.append(loss.data.numpy())#每个批次损失装载在列表
    
    # 每隔100步输出一下损失值（loss）
    if i % 100==0:
        losses.append(np.mean(batch_loss))#每一百次迭代的损失均值装载在losses
        print(i, np.mean(batch_loss))
# 打印输出损失值
fig = plt.figure(figsize=(10, 7))
plt.plot(np.arange(len(losses))*100,losses, 'o-')#绘制损失曲线
plt.xlabel('epoch')
plt.ylabel('MSE')

3.2.使用pytorch已有网络框架

input_size = features.shape[1]
hidden_size = 10
output_size = 1
batch_size = 128
neu = torch.nn.Sequential(
    torch.nn.Linear(input_size, hidden_size),
    torch.nn.Sigmoid(),
    torch.nn.Linear(hidden_size, output_size),
)
cost = torch.nn.MSELoss()
optimizer = torch.optim.SGD(neu.parameters(), lr = 0.01)
losses = []
for i in range(1000):
    # 每128个样本点被划分为一个撮，在循环的时候一批一批地读取
    batch_loss = []
    # start和end分别是提取一个batch数据的起始和终止下标
    for start in range(0, len(X), batch_size):
        end = start + batch_size if start + batch_size < len(X) else len(X)
        xx = torch.tensor(X[start:end], dtype = torch.float, requires_grad = True)
        yy = torch.tensor(Y[start:end], dtype = torch.float, requires_grad = True)
        predict = neu(xx)
        loss = cost(predict, yy)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        batch_loss.append(loss.data.numpy())
    
    # 每隔100步输出一下损失值（loss）
    if i % 100==0:
        losses.append(np.mean(batch_loss))
        print(i, np.mean(batch_loss))

4.测试

targets = test_targets['cnt'] #读取测试集的cnt数值
targets = targets.values.reshape([len(targets),1]) #将数据转换成合适的tensor形式
targets = targets.astype(float) #保证数据为实数

x = torch.tensor(test_features.values, dtype =  torch.double, requires_grad = True)#使用3.1搭建网络，则dtype= torch.double；使用3.2搭建3.2搭建网络，则dtype = torch.float
y = torch.tensor(targets, dtype = torch.float, requires_grad = True)

print(x[:10])
# 用神经网络进行预测
predict = neu(x)
predict = predict.data.numpy()
print((predict * std + mean)[:10])
# 将后21天的预测数据与真实数据画在一起并比较
# 横坐标轴是不同的日期，纵坐标轴是预测或者真实数据的值
fig, ax = plt.subplots(figsize = (10, 7))
mean, std = scaled_features['cnt']
ax.plot(predict * std + mean, label='Prediction', linestyle = '--')
ax.plot(targets * std + mean, label='Data', linestyle = '-')
ax.legend()
ax.set_xlabel('Date-time')
ax.set_ylabel('Counts')
# 对横坐标轴进行标注
dates = pd.to_datetime(rides.loc[test_data.index]['dteday'])
dates = dates.apply(lambda d: d.strftime('%b %d'))
ax.set_xticks(np.arange(len(dates))[12::24])
_ = ax.set_xticklabels(dates[12::24], rotation=45)

5.结果图