用pytorch实现一个简单的线性回归模型

我们使用pytorch实现一个简单的线性回归模型。(我是在jupyter notebook下做的，建议在这个环境下写代码)

首先我们先定义数据集：

#数据集中的样本值
x = [35.7, 55.9, 58.2, 81.9, 56.3, 48.9, 33.9, 21.8, 48.4, 60.4, 68.4]


#数据集中的标签
y = [0.5,  14.0, 15.0, 28.0, 11.0,  8.0,  3.0, -4.0,  6.0, 13.0, 21.0]


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

创建线性模型与损失函数，在这里我们的损失函数是平方均方误差函数：

def model(x, w, b):
    return w * x + b

def loss_fn(y_pre, y):
    squared_diffs = (y_pre - y) ** 2
    return squared_diffs.mean()

做分析(非必须)：

w = torch.ones(1)
b = torch.zeros(1)

y_pre = model(x, w, b)
y_pre

loss = loss_fn(y_pre, y)
loss

之后我们要做梯度下降，做梯度下降，我就需要对w和b求偏导。

def loss_fn(y_pre, y):
    squared_diffs = (y_pre - y) ** 2
    return squared_diffs.mean()

def dloss_fn(y_pre, y):
    dsq_diffs = 2 * (y_pre - y)
    return dsq_diffs

def dmodel_dw(x, w, b):
    return x

def dmodel_db(x, w, b):
    return 1.0

def grad_fn(x, y, y_pre, w, b):
    dloss_dw = dloss_fn(y_pre, y) * dmodel_dw(x, w, b)
    dloss_db = dloss_fn(y_pre, y) * dmodel_db(x, w, b)
    return torch.stack([dloss_dw.mean(), dloss_db.mean()])

我们写最后训练函数：

def training_loop(n_epochs, learning_rate, params, x, y):
    for epoch in range(1, n_epochs + 1):
        w, b   = params
        y_pre  = model(x, w, b)
        loss   = loss_fn(y_pre, y)
        grad   = grad_fn(x, y, y_pre, w, b)
        params = params - learning_rate * grad
        print('Epoch %d, Loss %f' % (epoch, float(loss)))
    return params

最后的训练效果：

全部代码示例：

import torch

#数据集中的样本值
x = [35.7, 55.9, 58.2, 81.9, 56.3, 48.9, 33.9, 21.8, 48.4, 60.4, 68.4]


#数据集中的标签
y = [0.5,  14.0, 15.0, 28.0, 11.0,  8.0,  3.0, -4.0,  6.0, 13.0, 21.0]


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


def model(x, w, b):
    return w * x + b

def loss_fn(y_pre, y):
    squared_diffs = (y_pre - y) ** 2
    return squared_diffs.mean()

w = torch.ones(1)
b = torch.zeros(1)

y_pre = model(x, w, b)
y_pre

loss = loss_fn(y_pre, y)
loss

def loss_fn(y_pre, y):
    squared_diffs = (y_pre - y) ** 2
    return squared_diffs.mean()

def dloss_fn(y_pre, y):
    dsq_diffs = 2 * (y_pre - y)
    return dsq_diffs

def dmodel_dw(x, w, b):
    return x

def dmodel_db(x, w, b):
    return 1.0

def grad_fn(x, y, y_pre, w, b):
    dloss_dw = dloss_fn(y_pre, y) * dmodel_dw(x, w, b)
    dloss_db = dloss_fn(y_pre, y) * dmodel_db(x, w, b)
    return torch.stack([dloss_dw.mean(), dloss_db.mean()])


def training_loop(n_epochs, learning_rate, params, x, y):
    for epoch in range(1, n_epochs + 1):
        w, b   = params
        y_pre  = model(x, w, b)
        loss   = loss_fn(y_pre, y)
        grad   = grad_fn(x, y, y_pre, w, b)
        params = params - learning_rate * grad
        print('Epoch %d, Loss %f' % (epoch, float(loss)))
    return params

training_loop(
    n_epoches     = 20,
    learning_rate = 1e-4,
    params        = torch.tensor([1.0, 0.0]),
    x             = x,
    y             = y
)

training_loop(
    n_epoches     = 20,
    learning_rate = 1e-2,
    params        = torch.tensor([1.0, 0.0]),
    x             = x,
    y             = y
)

 

参考资料：

《Deep Learning with PyTorch》Essential Excepts. Eli Stevens, Luca Antiga.