Follow Tang Yudi pytorch The second day of study , I'm going to use it today pytorch Build a neural network to predict the temperature .

One 、 Data presentation

        1.1 Reading data ：

        Use pandas Inside read_csv Method to read csv Data file and display ：

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import torch
import torch.optim as optim
import warnings
import datetime
from sklearn import preprocessing
warnings.filterwarnings("ignore")

features = pd.read_csv('')
print(features.head())                    # View the data 
print(" Data dimension ：",features.shape)    # Print data dimension 

        The data is shown in the figure ：

        The specific meaning of the data is ：

        year、month、day、week Respectively represent the day of the week ;

        temp_1 It means the highest temperature yesterday ,temp_2 Indicates the highest temperature of the day before yesterday ;average In history , The average temperature on this day of each year ;actual Represents the real maximum temperature of the day , This is equivalent to the tag value ;friend Indicates the possible value of your friend's guess , Leave it here for the time being .

        1.2 Data preprocessing

        First, turn the data into datetime Required standard format ：

years = features['year']            # Get years 
months = features['month']         # Get a month 
days = features['day']             # Day by day 
# To datetime The required standard format 
dates = [str(int(year)) + '-' + str(int(month)) + '-' + str(int(day)) for year,month,day in zip(years,months,days)]
dates = [datetime.datetime.strptime(date,'%Y-%m-%d') for date in dates]

        Here, draw an image to see what it looks like ：

# Drawing data images 
plt.style.use('fivethirtyeight')    # Specify the default style 
fig, ((ax1, ax2), (ax3,ax4)) = plt.subplots(nrows=2, ncols=2, figsize = (10,10))    # Setting up layout 
fig.autofmt_xdate(rotation=45)

ax1.plot(dates,features['actual'])  # Draw labels 
ax1.set_xlabel(''); ax1.set_ylabel('Temperature'); ax1.set_title('Max Temp')

ax2.plot(dates,features['temp_1'])  # Draw the temperature yesterday 
ax2.set_xlabel(''); ax2.set_ylabel('Temperature'); ax2.set_title('Previous Max Temp')

ax3.plot(dates,features['temp_2'])  # Draw the temperature of the day before yesterday 
ax3.set_xlabel(''); ax3.set_ylabel('Temperature'); ax3.set_title('Two Days Prior Max Temp')

ax4.plot(dates,features['friend'])  # Draw friend predictions 
ax4.set_xlabel(''); ax4.set_ylabel('Temperature'); ax4.set_title('Friend')

plt.tight_layout(pad=2)

        Because of week One column is of string type , Here we use unique coding to transform it , It is worth noting that , What we use here is pandas Inside get_dummies Method , You can directly read the data in and then determine which strings are encoded ：

# Hot coding alone 
features = pd.get_dummies(features)
print(features.head(5))

        Extract labels separately ：

# label 
labels = np.array(features['actual'])

# Remove labels from features 
features = features.drop('actual',axis=1)

# Keep your name separate , In case of future trouble 
feature_list = list(features.columns)

# Convert to appropriate format 
features = np.array(features)

# Standardization , After execution, the floating range of the value will become smaller 
input_features = preprocessing.StandardScaler().fit_transform(features)

Two 、 Build a network model  

        2.1 To build the network （ A more complicated method ）

        Convert data to tensor Format ：

x = torch.tensor(input_features,dtype=float)
y = torch.tensor(labels,dtype=float)

        Build hidden layer （ here bias The number of is the same as the characteristic number of the corresponding layer ）, Set the learning rate and loss function ：

# Weight parameter initialization 
weights = torch.randn((14,128), dtype=float, requires_grad=True)        #14 A feature is converted to 128 A hidden layer feature 
biases = torch.randn(128, dtype=float, requires_grad=True)              # Each neuron has a bias 
weights2 = torch.randn((128,1), dtype=float, requires_grad=True)
biases2 = torch.randn(1,dtype=float,requires_grad=True)

learning_rate = 0.001
losses = []

        Training ：

for i in range(1000):
    # Calculate hidden layer 
    hidden = x.mm(weights) + biases
    # Add activation function 
    hidden = torch.relu(hidden)
    # Predicted results 
    prediction = hidden.mm(weights2) + biases2
    # Calculate the loss 
    loss = torch.mean((prediction - y)**2)
    losses.append(loss.data.numpy())                # The above is the forward propagation process 

    # Print loss value 
    if i % 100 == 0:
        print('loss:',loss)
    # Back propagation , After back propagation, we get the gradient value of a parameter , After that, you need to further update the parameters 
    loss.backward()

    # Update parameters , Multiply it in the opposite direction of the gradient , there - Represents the opposite direction 
    weights.data.add_(- learning_rate * weights.grad.data)
    biases.data.add_(- learning_rate * biases.grad.data)
    weights.data.add_(- learning_rate * weights2.grad.data)
    biases2.data.add_(- learning_rate * biases2.grad.data)

    # Every iteration needs to clear the gradient 
    weights.grad.data.zero_()
    biases.grad.data.zero_()
    weights2.grad.data.zero_()
    biases2.grad.data.zero_()

         2.2 A simpler way

input_size = input_features.shape[1]
hidden_size = 128
output_size = 1
batch_size = 16
my_nn = torch.nn.Sequential(
    torch.nn.Linear(input_size,hidden_size),
    torch.nn.Sigmoid(),
    torch.nn.Linear(hidden_size,output_size),
)
cost = torch.nn.MSELoss(reduction='mean')
optimizer = torch.optim.Adam(my_nn.parameters(),lr=0.001)
losses = []
# Training network 
for i in range(1000):
    batch_loss = []
    # use MINI-Batch Methods to train 
    for start in range(0, len(input_features), batch_size):
        end = start + batch_size if start + batch_size < len(input_features) else len(input_features)
        xx = torch.tensor(input_features[start:end], dtype=torch.float,requires_grad=True)
        yy = torch.tensor(labels[start:end],dtype=torch.float,requires_grad=True)
        prediction = my_nn(xx)
        loss = cost(prediction,yy)
        optimizer.zero_grad()
        loss.backward(retain_graph = True)
        optimizer.step()
        batch_loss.append(loss.data.numpy())

    if i % 100 == 0:
        losses.append(np.mean(batch_loss))
        print(i,np.mean(batch_loss))