简单的对二维数据进行线性拟合

简单的对二维数据进行线性拟合

环境为CUDA10.1，torch1.7.1+cu101， python3.7

数据生成

做一个一维数据，基本原理是给一个线性函数加点噪声

import numpy as np
import random
import csv
import pandas as pd
import torch
import matplotlib.pyplot as plt
a = torch.rand(100, 1)
data = np.zeros([100, 2], dtype=int)
for i in range(100):
    data[i, 0] = i
    data[i, 1] = random.uniform(0, 10) + 0.5 * i
print("x = ", data[:, 0])
print("y = ", data[:, 1])
# 将数据存为CSV格式
with open("data_CSV/liner_data.csv", "w", newline="") as f:
    csv_writer = csv.writer(f)
    csv_writer.writerow(["x", "y"])
    for i in range(100):
        csv_writer.writerow([data[i, 0], data[i, 1]])
# 读取数据并打印
csv_reader = pd.read_csv("data_CSV/liner_data.csv")
print("csv_reader = ", csv_reader)
plt.figure(figsize=(30, 10))
plt.scatter(data[:, 0], data[:, 1], color='red')
plt.show()

线性拟合

import numpy as np
import csv
import pandas as pd
import torch
import torch.nn as nn
import matplotlib.pyplot as plt
from torch.autograd import Variable

# 加载数据
csv_reader = pd.read_csv("data_CSV/liner_data.csv")
# print("csv_reader = ", csv_reader)
X = torch.empty(100, 1)
Y = torch.empty(100, 1)
for i in range(100):
    X[i, 0] = torch.tensor(csv_reader["x"][i], dtype=torch.float)
for i in range(100):
    Y[i, 0] = torch.tensor(csv_reader["y"][i], dtype=torch.float)

# 定义一个模型
class Liner_model(nn.Module):
    def __init__(self):
        super(Liner_model, self).__init__()
        self.linear = nn.Linear(1, 1)

    def forward(self, input_data):
        # y = weight * input_data + bias
        return self.linear(input_data)
#
#
lm = Liner_model()
# print("lm.parameters() = ", lm.parameters())
# 定义损失函数和梯度下降方法
criterion = nn.MSELoss()
optim = torch.optim.SGD(lm.parameters(), lr=1e-5)
# print("optim = ", optim)
Predict = torch.empty(100, 1)
for epoch in range(100):
    predict = lm(Variable(X[epoch]))
    Predict[epoch, 0] = torch.tensor(predict, dtype=torch.float)
    loss = criterion(predict, Variable(Y[epoch]))
    if epoch and epoch % 1 == 0:
        print("Loss:{:.3f}".format(loss.item()))
    optim.zero_grad()
    loss.backward()
    optim.step()
lm.eval()
predict = lm(Variable(X))
predict = predict.data.numpy()
# 查看数据
plt.figure(figsize=(30, 10))
fig, ax1 = plt.subplots()
fig.suptitle("Liner sample")
ax1.set_xlabel("x")
ax1.set_ylabel("y")
plt.scatter(X, Y, color='red')   # 红色为原始数据
plt.scatter(X, predict, color='blue') # 蓝色为最终预测模型
plt.scatter(X, Predict, color='black') # 黑色为训练过程中模型预测的变化
plt.savefig("data_CSV/liner_train.jpg")
plt.show()

总结

实际操作中，学习率、损失函数、梯度下降法可以更换康康效果。