Pytorch学习之多维线性回归

达达

声明库函数

import torch
import numpy as np
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
import random

定义多项式特征

#定义最高次项系数
n=3

def make_features(x):
    """构建一个[x,x^2,x^3]矩阵特征的实例i.e"""
    # unqueeze(1)为增加一维
    x = x.unsqueeze(1)
    return torch.cat([x ** i for i in range(1,n+1)],1)

定义真实函数

# y = 0.9 + 0.5x + 3x^2 + 2.4x^3
W_target = torch.FloatTensor([0.5,3,2.4]).unsqueeze(1)
b_target = torch.FloatTensor([0.9])

函数近似值获取

def f(x):
    """近似功能"""
    return x.mm(W_target)+b_target[0]

获得batch

def get_batch(batch_size = 32, random = None) :
    if random is None :
        random = torch.randn(batch_size)
    batch_size = random.size()[0]
    x = make_features(random)
    y = f(x)
    if torch.cuda.is_available() :
        return Variable(x).cuda(), Variable(y).cuda()
    else :
        return Variable(x), Variable(y)

定义多项式模型

# 定义模型
class ploy_model(nn.Module):
    def __init__(self):
        super().__init__()
        self.poly = nn.Linear(n,1)

    def forward(self,x):
        out = self.poly(x)
        return out

if torch.cuda.is_available():
    model = ploy_model().cuda()
else:
    model = ploy_model()

定义损失函数和优化器

# 损失函数和优化器
criterion = nn.MSELoss()
optimizer = optim.SGD(model.parameters(),lr=1e-3)

模型训练

epoch = 0
while True:
    # 获得数据
    batch_x,batch_y = get_batch()
    # 向前传播
    output = model(batch_x)
    loss = criterion(output,batch_y)
    print_loss = loss.data[0]
    # 重置梯度
    optimizer.zero_grad()
    # 后向传播
    loss.backward()
    # 更新参数
    optimizer.step()
    epoch +=1
    if print_loss < 1e-3:
        break

print("the number of epoches :", epoch)

函数预测

# 定义函数输出形式
def func_format(weight, bias, n):
    func = ''
    for i in range(n, 0, -1):
        func += ' {:.2f} * x^{} +'.format(weight[i - 1], i)
    return 'y =' + func + ' {:.2f}'.format(bias[0])

predict_weight = model.poly.weight.data.numpy().flatten()
predict_bias = model.poly.bias.data.numpy().flatten()
print('predicted function :', func_format(predict_weight, predict_bias,3))
real_W = W_target.numpy().flatten()
real_b = b_target.numpy().flatten()
print('real      function :', func_format(real_W, real_b,3))

结果

predicted function : y = 2.39 * x^3 + 2.99 * x^2 + 0.54 * x^1 + 0.93
real function : y = 2.40 * x^3 + 3.00 * x^2 + 0.50 * x^1 + 0.90

函数绘图

import matplotlib.pyplot as plt
import numpy as np
x = [random.randint(-200, 200) * 0.01 for i in range(20)]
x = np.array(sorted(x))
feature_x, y = get_batch(random = torch.from_numpy(x).float())
y = y.data.numpy()
plt.plot(x, y, 'ro', label='Original data')

model.eval()
x_sample = np.arange(-2, 2, 0.01)
x, y = get_batch(random = torch.from_numpy(x_sample).float())
y = model(x)
y_sample = y.data.numpy()
plt.plot(x_sample, y_sample, label = 'Fitting Line')
plt.show()

image.png

最后编辑于：2019.01.06 15:26:27