PyTorchニューラルネットワークの温度予測

概要概要

特定のケースの説明はここでは繰り返されません。同じデータセットが、機械学習のランダムフォレストモデルで説明されています。

ガイドパッケージ

import numpy as np
import pandas as pd
import datetime
import matplotlib.pyplot as plt
from pandas.plotting import register_matplotlib_converters
from sklearn.preprocessing import StandardScaler
import torch

読み取ったデータ

# ------------------1. 数据读取------------------

# 读取数据
data = pd.read_csv("temps.csv")

# 看看数据长什么样子
print(data.head())

# 查看数据维度
print("数据维度:", data.shape)

# 产看数据类型
print("数据类型:", type(data))

输出结果:
   year  month  day  week  temp_2  temp_1  average  actual  friend
0  2016      1    1   Fri      45      45     45.6      45      29
1  2016      1    2   Sat      44      45     45.7      44      61
2  2016      1    3   Sun      45      44     45.8      41      56
3  2016      1    4   Mon      44      41     45.9      40      53
4  2016      1    5  Tues      41      40     46.0      44      41
数据维度: (348, 9)
数据类型: <class 'pandas.core.frame.DataFrame'>

データの前処理

# ------------------2. 数据预处理------------------

# datetime 格式
dates = pd.PeriodIndex(year=data["year"], month=data["month"], day=data["day"], freq="D").astype(str)
dates = [datetime.datetime.strptime(date, "%Y-%m-%d") for date in dates]
print(dates[:5])

# 编码转换
data = pd.get_dummies(data)
print(data.head())

# 画图
plt.style.use("fivethirtyeight")
register_matplotlib_converters()

# 标签
labels = np.array(data["actual"])

# 取消标签
data = data.drop(["actual"], axis= 1)
print(data.head())

# 保存一下列名
feature_list = list(data.columns)

# 格式转换
data_new = np.array(data)

data_new  = StandardScaler().fit_transform(data_new)
print(data_new[:5])

输出结果:
[datetime.datetime(2016, 1, 1, 0, 0), datetime.datetime(2016, 1, 2, 0, 0), datetime.datetime(2016, 1, 3, 0, 0), datetime.datetime(2016, 1, 4, 0, 0), datetime.datetime(2016, 1, 5, 0, 0)]
   year  month  day  temp_2  ...  week_Sun  week_Thurs  week_Tues  week_Wed
0  2016      1    1      45  ...         0           0          0         0
1  2016      1    2      44  ...         0           0          0         0
2  2016      1    3      45  ...         1           0          0         0
3  2016      1    4      44  ...         0           0          0         0
4  2016      1    5      41  ...         0           0          1         0

[5 rows x 15 columns]
   year  month  day  temp_2  ...  week_Sun  week_Thurs  week_Tues  week_Wed
0  2016      1    1      45  ...         0           0          0         0
1  2016      1    2      44  ...         0           0          0         0
2  2016      1    3      45  ...         1           0          0         0
3  2016      1    4      44  ...         0           0          0         0
4  2016      1    5      41  ...         0           0          1         0

[5 rows x 14 columns]
[[ 0.         -1.5678393  -1.65682171 -1.48452388 -1.49443549 -1.3470703
  -1.98891668  2.44131112 -0.40482045 -0.40961596 -0.40482045 -0.40482045
  -0.41913682 -0.40482045]
 [ 0.         -1.5678393  -1.54267126 -1.56929813 -1.49443549 -1.33755752
   0.06187741 -0.40961596 -0.40482045  2.44131112 -0.40482045 -0.40482045
  -0.41913682 -0.40482045]
 [ 0.         -1.5678393  -1.4285208  -1.48452388 -1.57953835 -1.32804474
  -0.25855917 -0.40961596 -0.40482045 -0.40961596  2.47023092 -0.40482045
  -0.41913682 -0.40482045]
 [ 0.         -1.5678393  -1.31437034 -1.56929813 -1.83484692 -1.31853195
  -0.45082111 -0.40961596  2.47023092 -0.40961596 -0.40482045 -0.40482045
  -0.41913682 -0.40482045]
 [ 0.         -1.5678393  -1.20021989 -1.8236209  -1.91994977 -1.30901917
  -1.2198689  -0.40961596 -0.40482045 -0.40961596 -0.40482045 -0.40482045
   2.38585576 -0.40482045]]

ネットワークモデルを構築する

# ------------------3. 构建网络模型------------------

x = torch.tensor(data_new)
y = torch.tensor(labels)

# 权重参数初始化
weights1 = torch.randn((14,128), dtype=float, requires_grad= True)
biases1 = torch.randn(128, dtype=float, requires_grad= True)
weights2 = torch.randn((128,1), dtype=float, requires_grad= True)
biases2 = torch.randn(1, dtype=float, requires_grad= True)

learning_rate = 0.001
losses = []

for i in range(1000):
    # 计算隐层
    hidden = x.mm(weights1) + biases1
    # 加入激活函数
    hidden = torch.relu(hidden)
    # 预测结果
    predictions = hidden.mm(weights2) + biases2
    # 计算损失
    loss = torch.mean((predictions - y) ** 2)

    # 打印损失值
    if i % 100 == 0:
        print("loss:", loss)
    # 反向传播计算
    loss.backward()

    # 更新参数
    weights1.data.add_(-learning_rate * weights1.grad.data)
    biases1.data.add_(-learning_rate * biases1.grad.data)
    weights2.data.add_(-learning_rate * weights2.grad.data)
    biases2.data.add_(-learning_rate * biases2.grad.data)

    # 每次迭代清空
    weights1.grad.data.zero_()
    biases1.grad.data.zero_()
    weights2.grad.data.zero_()
    biases2.grad.data.zero_()

输出结果:
loss: tensor(4746.8598, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(156.5691, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(148.9419, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(146.1035, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(144.5652, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(143.5376, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(142.7823, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(142.2151, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(141.7770, dtype=torch.float64, grad_fn=<MeanBackward0>)
loss: tensor(141.4294, dtype=torch.float64, grad_fn=<MeanBackward0>)

データの視覚化

# ------------------4. 数据可视化------------------

def graph1():
    # 创建子图
    f, ax = plt.subplots(2, 2, figsize=(10, 10))

    # 标签值
    ax[0, 0].plot(dates, labels, color="#ADD8E6")
    ax[0, 0].set_xticks([""])
    ax[0, 0].set_ylabel("Temperature")
    ax[0, 0].set_title("Max Temp")

    # 昨天
    ax[0, 1].plot(dates, data["temp_1"], color="#87CEFA")
    ax[0, 1].set_xticks([""])
    ax[0, 1].set_ylabel("Temperature")
    ax[0, 1].set_title("Previous Max Temp")

    # 前天
    ax[1, 0].plot(dates, data["temp_2"], color="#00BFFF")
    ax[1, 0].set_xticks([""])
    ax[1, 0].set_xlabel("Date")
    ax[1, 0].set_ylabel("Temperature")
    ax[1, 0].set_title("Two Days Prior Max Temp")

    # 朋友
    ax[1, 1].plot(dates, data["friend"], color="#1E90FF")
    ax[1, 1].set_xticks([""])
    ax[1, 1].set_xlabel("Date")
    ax[1, 1].set_ylabel("Temperature")
    ax[1, 1].set_title("Friend Estimate")

    plt.show()

出力結果：

完全なコード

import numpy as np
import pandas as pd
import datetime
import matplotlib.pyplot as plt
from pandas.plotting import register_matplotlib_converters
from sklearn.preprocessing import StandardScaler
import torch


# ------------------1. 数据读取------------------

# 读取数据
data = pd.read_csv("temps.csv")

# 看看数据长什么样子
print(data.head())

# 查看数据维度
print("数据维度:", data.shape)

# 产看数据类型
print("数据类型:", type(data))

# ------------------2. 数据预处理------------------

# datetime 格式
dates = pd.PeriodIndex(year=data["year"], month=data["month"], day=data["day"], freq="D").astype(str)
dates = [datetime.datetime.strptime(date, "%Y-%m-%d") for date in dates]
print(dates[:5])

# 编码转换
data = pd.get_dummies(data)
print(data.head())

# 画图
plt.style.use("fivethirtyeight")
register_matplotlib_converters()

# 标签
labels = np.array(data["actual"])

# 取消标签
data = data.drop(["actual"], axis= 1)
print(data.head())

# 保存一下列名
feature_list = list(data.columns)

# 格式转换
data_new = np.array(data)

data_new  = StandardScaler().fit_transform(data_new)
print(data_new[:5])

# ------------------3. 构建网络模型------------------

x = torch.tensor(data_new)
y = torch.tensor(labels)

# 权重参数初始化
weights1 = torch.randn((14,128), dtype=float, requires_grad= True)
biases1 = torch.randn(128, dtype=float, requires_grad= True)
weights2 = torch.randn((128,1), dtype=float, requires_grad= True)
biases2 = torch.randn(1, dtype=float, requires_grad= True)

learning_rate = 0.001
losses = []

for i in range(1000):
    # 计算隐层
    hidden = x.mm(weights1) + biases1
    # 加入激活函数
    hidden = torch.relu(hidden)
    # 预测结果
    predictions = hidden.mm(weights2) + biases2
    # 计算损失
    loss = torch.mean((predictions - y) ** 2)

    # 打印损失值
    if i % 100 == 0:
        print("loss:", loss)
    # 反向传播计算
    loss.backward()

    # 更新参数
    weights1.data.add_(-learning_rate * weights1.grad.data)
    biases1.data.add_(-learning_rate * biases1.grad.data)
    weights2.data.add_(-learning_rate * weights2.grad.data)
    biases2.data.add_(-learning_rate * biases2.grad.data)

    # 每次迭代清空
    weights1.grad.data.zero_()
    biases1.grad.data.zero_()
    weights2.grad.data.zero_()
    biases2.grad.data.zero_()

# ------------------4. 数据可视化------------------

def graph1():
    # 创建子图
    f, ax = plt.subplots(2, 2, figsize=(10, 10))

    # 标签值
    ax[0, 0].plot(dates, labels, color="#ADD8E6")
    ax[0, 0].set_xticks([""])
    ax[0, 0].set_ylabel("Temperature")
    ax[0, 0].set_title("Max Temp")

    # 昨天
    ax[0, 1].plot(dates, data["temp_1"], color="#87CEFA")
    ax[0, 1].set_xticks([""])
    ax[0, 1].set_ylabel("Temperature")
    ax[0, 1].set_title("Previous Max Temp")

    # 前天
    ax[1, 0].plot(dates, data["temp_2"], color="#00BFFF")
    ax[1, 0].set_xticks([""])
    ax[1, 0].set_xlabel("Date")
    ax[1, 0].set_ylabel("Temperature")
    ax[1, 0].set_title("Two Days Prior Max Temp")

    # 朋友
    ax[1, 1].plot(dates, data["friend"], color="#1E90FF")
    ax[1, 1].set_xticks([""])
    ax[1, 1].set_xlabel("Date")
    ax[1, 1].set_ylabel("Temperature")
    ax[1, 1].set_title("Friend Estimate")

    plt.show()


if __name__ == "__main__":
    graph1()