构造一个小的回归数据集:
生成 150 个带噪音的样本,其中 100 个训练样本,50 个测试样本,并打印出训练数据的可视化分布。
import numpy as np
import torch
import pandas as pd
import matplotlib.pyplot as plt
def creat_data(w,b,num_examples):
x = torch.normal(0,1,(num_examples,len(w)))
'''
标准正态分布,对应于np.random.normal(loc=0, scale=1, size=shape)
参数loc(float):正态分布的均值,对应着这个分布的中心。loc=0说明这一个以Y轴为对称轴的正态分布,
参数scale(float):正态分布的标准差,对应分布的宽度,scale越大,正态分布的曲线越矮胖,scale越小,曲线越高瘦。
参数size(int 或者整数元组):输出的值赋在shape里,默认为None
'''
y = np.matmul(x, w) + b
# numpy.matmul 函数返回两个数组的矩阵乘积。当两个数组都是二维数组的时候,就是数学上的两个矩阵的乘积。
return x,y.reshape(-1,1)
# numpy.reshape(-1,1),将原来的数据,改为1列(注意reshape的1),但是行数不确定,numpy库自己计算(注意reshape的-1)。
# 添加噪声
def get_noise(loc,scale,y):
noise = torch.normal(loc,scale,y.shape)
return y + noise
true_w = [1.2]
true_b = 0.5
features, labels = creat_data(true_w, true_b, 150)
# 增加噪声
labels = get_noise(0,0.1,labels)
train_features = features[0:100]
train_labels = labels[0:100]
test_features = features[100:150]
test_labels = labels[100:150]
# 图像展示
plt.figure(1)
plt.scatter(train_features,train_labels,c='r')
plt.show()
import torch
from op import Op
torch.manual_seed(10) # 设置随机种子
# 线性算子
class Linear(Op):
def __init__(self, input_size):
"""
输入:
- input_size:模型要处理的数据特征向量长度
"""
self.input_size = input_size
# 模型参数
self.params = {}
self.params['w'] = torch.randn(self.input_size, 1)
self.params['b'] = torch.zeros(size=[1], dtype=torch.float32)
def __call__(self, X):
return self.forward(X)
# 前向函数
def forward(self, X):
"""
输入:
- X: tensor, shape=[N,D]
注意这里的X矩阵是由N个x向量的转置拼接成的,与原教材行向量表示方式不一致
输出:
- y_pred: tensor, shape=[N]
"""
N, D = X.shape
if self.input_size == 0:
return torch.full(shape=[N, 1], fill_value=self.params['b'])
assert D == self.input_size # 输入数据维度合法性验证
# 使用paddle.matmul计算两个tensor的乘积
y_pred = torch.matmul(X, self.params['w']) + self.params['b']
return y_pred
input_size = 3
N = 2
X = torch.randn(N, input_size) # 生成2个维度为3的数据
model = Linear(input_size)
y_pred = model(X)
print("y_pred:", y_pred) # 输出结果的个数也是2个
y_pred: tensor([[1.8529],
[0.6011]])
回归任务中常用的评估指标是均方误差
均方误差(mean-square error, MSE)是反映估计量与被估计量之间差异程度的一种度量。
import torch
def mean_squared_error(y_true, y_pred):
"""
输入:
- y_true: tensor,样本真实标签
- y_pred: tensor, 样本预测标签
输出:
- error: float,误差值
"""
assert y_true.shape[0] == y_pred.shape[0]
error = torch.mean(torch.square(y_true - y_pred))
return error
# 构造一个简单的样例进行测试:[N,1], N=2
y_true = torch.tensor([[-0.2], [4.9]], dtype=torch.float32)
y_pred = torch.tensor([[1.3], [2.5]], dtype=torch.float32)
error = mean_squared_error(y_true=y_true, y_pred=y_pred).item()
print("error:", error)
error: 4.005000114440918
进程已结束,退出代码为 0
思考:没有除2是不影响的,代码中没有除以二,损失函数公式中含有平方,在之后对其进行求导,会多出一个二,所以公式中除以二以简化求导后的结果
经验风险 ( Empirical Risk ),即在训练集上的平均损失。
def optimizer_lsm(model, X, y, reg_lambda=0):
'''
输入:
- model: 模型
- X: tensor, 特征数据,shape=[N,D]
- y: tensor,标签数据,shape=[N]
- reg_lambda: float, 正则化系数,默认为0
输出:
- model: 优化好的模型
'''
N, D = X.shape
# 对输入特征数据所有特征向量求平均
x_bar_tran = torch.mean(X,0).T
# 求标签的均值,shape=[1]
y_bar = torch.mean(y)
# torch.subtract通过广播的方式实现矩阵减向量
x_sub = torch.subtract(X, x_bar_tran)
# 使用torch.all判断输入tensor是否全0
if torch.all(x_sub == 0):
model.params['b'] = y_bar
model.params['w'] = torch.zeros([D])
return model
# torch.inverse求方阵的逆
tmp = torch.inverse(torch.matmul(x_sub.T, x_sub) +
reg_lambda * torch.eye(D))
w = torch.matmul(torch.matmul(tmp, x_sub.T), (y - y_bar))
b = y_bar - torch.matmul(x_bar_tran, w)
model.params['b'] = b
model.params['w'] = torch.squeeze(w, -1)
return model
思考:
1.省略1/N不影响,1为N维的全一向量。
2.最小二乘法是一种数学优化技术。它通过最小化误差的平方和寻找数据的最佳函数匹配。利用它可以求得的数据,使这些数据与实际数据之间误差的平方和最小
2.2.5 模型训练
在准备了数据、模型、损失函数和参数学习的实现之后,开始模型的训练。
在回归任务中,模型的评价指标和损失函数一致,都为均方误差。
通过上文实现的线性回归类来拟合训练数据,并输出模型在训练集上的损失。
input_size = 1
model = Linear(input_size)
model = optimizer_lsm(model,X_train.reshape([-1,1]),y_train.reshape([-1,1]))
print("w_pred:", model.params['w'].item(), "b_pred: ", model.params['b'].item())
y_train_pred = model(X_train.reshape([-1,1])).squeeze()
train_error = mean_squared_error(y_true=y_train, y_pred=y_train_pred).item()
print("train error: ", train_error)
w_pred: 1.2271720170974731 b_pred: 0.37986236810684204
train error: 3.632181406021118
model_large = Linear(input_size)
model_large = optimizer_lsm(model_large,X_train_large.reshape([-1,1]),y_train_large.reshape([-1,1]))
print("w_pred large:",model_large.params['w'].item(), "b_pred large: ", model_large.params['b'].item())
y_train_pred_large = model_large(X_train_large.reshape([-1,1])).squeeze()
train_error_large = mean_squared_error(y_true=y_train_large, y_pred=y_train_pred_large).item()
print("train error large: ",train_error_large)
w_pred large: 1.1993682384490967 b_pred large: 0.5421561002731323
train error large: 3.9921984672546387
用训练好的模型预测一下测试集的标签,并计算在测试集上的损失。
y_test_pred = model(X_test.reshape([-1,1])).squeeze()
test_error = mean_squared_error(y_true=y_test, y_pred=y_test_pred).item()
print("test error: ",test_error)
test error: 4.306798458099365
y_test_pred_large = model_large(X_test.reshape([-1,1])).squeeze()
test_error_large = mean_squared_error(y_true=y_test, y_pred=y_test_pred_large).item()
print("test error large: ",test_error_large)
test error large: 4.420894622802734
构建训练和测试数据,其中:
# sin函数: sin(2 * pi * x)
def sin(x):
y = torch.sin(2 * math.pi * x)
return y
# 生成数据
func = sin
interval = (0,1)
train_num = 15
test_num = 10
noise = 0.5 #0.1
X_train, y_train = create_toy_data(func=func, interval=interval, sample_num=train_num, noise = noise)
X_test, y_test = create_toy_data(func=func, interval=interval, sample_num=test_num, noise = noise)
X_underlying = torch.linspace(interval[0],interval[1],100)
y_underlying = sin(X_underlying)
# 绘制图像
plt.rcParams['figure.figsize'] = (8.0, 6.0)
plt.scatter(X_train, y_train, facecolor="none", edgecolor='#e4007f', s=50, label="train data")
#plt.scatter(X_test, y_test, facecolor="none", edgecolor="r", s=50, label="test data")
plt.plot(X_underlying, y_underlying, c='#000000', label=r"$\sin(2\pi x)$")
plt.legend(fontsize='x-large')
plt.savefig('ml-vis2.pdf')
plt.show()
训练数样本 15 个,测试样本 10 个,高斯噪声标准差为 0.1,自变量范围为 (0,1)。
# 多项式转换
def polynomial_basis_function(x, degree=2):
"""
输入:
- x: tensor, 输入的数据,shape=[N,1]
- degree: int, 多项式的阶数
example Input: [[2], [3], [4]], degree=2
example Output: [[2^1, 2^2], [3^1, 3^2], [4^1, 4^2]]
注意:本案例中,在degree>=1时不生成全为1的一列数据;degree为0时生成形状与输入相同,全1的Tensor
输出:
- x_result: tensor
"""
if degree == 0:
# x = torch.ones(x.shape)
# x = x.to(torch.float32)
# return x
return torch.ones(x.shape)
x_tmp = x
x_result = x_tmp
for i in range(2, degree + 1):
x_tmp = torch.mul(x_tmp, x) # 逐元素相乘
x_result = torch.cat((x_result, x_tmp), dim=-1)
return x_result
# 简单测试
data = [[2], [3], [4]]
X = torch.tensor(data=data)
X = X.to(torch.float32)
degree = 3
transformed_X = polynomial_basis_function(X, degree=degree)
print("转换前:", X)
print("阶数为", degree, "转换后:", transformed_X)
转换前: tensor([[2.],
[3.],
[4.]])
阶数为 3 转换后: tensor([[ 2., 4., 8.],
[ 3., 9., 27.],
[ 4., 16., 64.]])
套用求解线性回归参数的方法来求解多项式回归参数
plt.rcParams['figure.figsize'] = (12.0, 8.0)
for i, degree in enumerate([0, 1, 3, 8]): # []中为多项式的阶数
model = Linear(degree)
X_train_transformed = polynomial_basis_function(X_train.reshape([-1, 1]), degree)
X_underlying_transformed = polynomial_basis_function(X_underlying.reshape([-1, 1]), degree)
model = optimizer_lsm(model, X_train_transformed, y_train.reshape([-1, 1])) # 拟合得到参数
y_underlying_pred = model(X_underlying_transformed).squeeze()
print(model.params)
# 绘制图像
plt.subplot(2, 2, i + 1)
plt.scatter(X_train, y_train, facecolor="none", edgecolor='#e4007f', s=50, label="train data")
plt.plot(X_underlying, y_underlying, c='#000000', label=r"$\sin(2\pi x)$")
plt.plot(X_underlying, y_underlying_pred, c='#f19ec2', label="predicted function")
plt.ylim(-2, 1.5)
plt.annotate("M={}".format(degree), xy=(0.95, -1.4))
# plt.legend(bbox_to_anchor=(1.05, 0.64), loc=2, borderaxespad=0.)
plt.legend(loc='lower left', fontsize='x-large')
plt.savefig('ml-vis3.pdf')
plt.show()
# 训练误差和测试误差
training_errors = []
test_errors = []
distribution_errors = []
# 遍历多项式阶数
for i in range(9):
model = Linear(i)
X_train_transformed = polynomial_basis_function(X_train.reshape([-1, 1]), i)
X_test_transformed = polynomial_basis_function(X_test.reshape([-1, 1]), i)
X_underlying_transformed = polynomial_basis_function(X_underlying.reshape([-1, 1]), i)
optimizer_lsm(model, X_train_transformed, y_train.reshape([-1, 1]))
y_train_pred = model(X_train_transformed).squeeze()
y_test_pred = model(X_test_transformed).squeeze()
y_underlying_pred = model(X_underlying_transformed).squeeze()
train_mse = mean_squared_error(y_true=y_train, y_pred=y_train_pred).item()
training_errors.append(train_mse)
test_mse = mean_squared_error(y_true=y_test, y_pred=y_test_pred).item()
test_errors.append(test_mse)
# distribution_mse = mean_squared_error(y_true=y_underlying, y_pred=y_underlying_pred).item()
# distribution_errors.append(distribution_mse)
print("train errors: \n", training_errors)
print("test errors: \n", test_errors)
# print ("distribution errors: \n", distribution_errors)
# 绘制图片
plt.rcParams['figure.figsize'] = (8.0, 6.0)
plt.plot(training_errors, '-.', mfc="none", mec='#e4007f', ms=10, c='#e4007f', label="Training")
plt.plot(test_errors, '--', mfc="none", mec='#f19ec2', ms=10, c='#f19ec2', label="Test")
# plt.plot(distribution_errors, '-', mfc="none", mec="#3D3D3F", ms=10, c="#3D3D3F", label="Distribution")
plt.legend(fontsize='x-large')
plt.xlabel("degree")
plt.ylabel("MSE")
plt.savefig('ml-mse-error.pdf')
plt.show()
机器学习方法流程包括数据集构建、模型构建、损失函数定义、优化器、模型训练、模型评价、模型预测等环节。
为了更方便地将上述环节规范化,我们将机器学习模型的基本要素封装成一个Runner类。
除上述提到的要素外,再加上模型保存、模型加载等功能。
Runner类的成员函数定义如下:
__init__函数:实例化Runner类,需要传入模型、损失函数、优化器和评价指标等;
train函数:模型训练,指定模型训练需要的训练集和验证集;
evaluate函数:通过对训练好的模型进行评价,在验证集或测试集上查看模型训练效果;
predict函数:选取一条数据对训练好的模型进行预测;
save_model函数:模型在训练过程和训练结束后需要进行保存;
load_model函数:调用加载之前保存的模型。
# Runner类
class Runner(object):
def __init__(self, model, optimizer, loss_fn, metric):
self.model = model # 模型
self.optimizer = optimizer # 优化器
self.loss_fn = loss_fn # 损失函数
self.metric = metric # 评估指标
# 模型训练
def train(self, train_dataset, dev_dataset=None, **kwargs):
pass
# 模型评价
def evaluate(self, data_set, **kwargs):
pass
# 模型预测
def predict(self, x, **kwargs):
pass
# 模型保存
def save_model(self, save_path):
pass
# 模型加载
def load_model(self, model_path):
pass
使用线性回归来对马萨诸塞州波士顿郊区的房屋进行预测。
实验流程主要包含如下5个步骤:
数据处理:包括数据清洗(缺失值和异常值处理)、数据集划分,以便数据可以被模型正常读取,并具有良好的泛化性;
模型构建:定义线性回归模型类;
训练配置:训练相关的一些配置,如:优化算法、评价指标等;
组装训练框架Runner:Runner用于管理模型训练和测试过程;
模型训练和测试:利用Runner进行模型训练和测试。
2.5.1 数据处理
2.5.1.1数据集
data = pd.read_csv(r'C:\Users\86181\Desktop\boston_house_prices.csv')
print(data.head())
CRIM ZN INDUS CHAS NOX ... RAD TAX PTRATIO LSTAT MEDV
0 0.00632 18.0 2.31 0 0.538 ... 1 296 15.3 4.98 24.0
1 0.02731 0.0 7.07 0 0.469 ... 2 242 17.8 9.14 21.6
2 0.02729 0.0 7.07 0 0.469 ... 2 242 17.8 4.03 34.7
3 0.03237 0.0 2.18 0 0.458 ... 3 222 18.7 2.94 33.4
4 0.06905 0.0 2.18 0 0.458 ... 3 222 18.7 5.33 36.2
[5 rows x 13 columns]
2.5.1.2 数据清洗
# 查看各字段缺失值统计情况
print(data.isna().sum())
CRIM 0
ZN 0
INDUS 0
CHAS 0
NOX 0
RM 0
AGE 0
DIS 0
RAD 0
TAX 0
PTRATIO 0
LSTAT 0
MEDV 0
dtype: int64
2.5.1.3 数据集划分
将数据集划分为两份:训练集和测试集,不包括验证集。
torch.manual_seed(10) # 随机种子
# 划分训练集和测试集
def train_test_split(X, y, train_percent=0.8):
n = len(X)
shuffled_indices = torch.randperm(n) # 返回一个数值在0到n-1、随机排列的1-D Tensor
train_set_size = int(n * train_percent)
train_indices = shuffled_indices[:train_set_size]
test_indices = shuffled_indices[train_set_size:]
X = X.values
y = y.values
X_train = X[train_indices]
y_train = y[train_indices]
X_test = X[test_indices]
y_test = y[test_indices]
return X_train, X_test, y_train, y_test
X = data.drop(['MEDV'], axis=1)
y = data['MEDV']
X_train, X_test, y_train, y_test = train_test_split(X, y) # X_train每一行是个样本,shape[N,D]
2.5.1.4 特征工程
为了消除纲量对数据特征之间影响,在模型训练前,需要对特征数据进行归一化处理,将数据缩放到[0, 1]区间内,使得不同特征之间具有可比性。
X_train = torch.tensor(X_train)
X_train = X_train.to(torch.float32)
X_test = torch.tensor(X_test)
X_train = X_train.to(torch.float32)
y_train = torch.tensor(y_train)
X_train = X_train.to(torch.float32)
y_test = torch.tensor(y_test)
X_train = X_train.to(torch.float32)
X_min = torch.min(X_train)
X_max = torch.max(X_train)
X_train = (X_train-X_min)/(X_max-X_min)
X_test = (X_test-X_min)/(X_max-X_min)
# 训练集构造
train_dataset = (X_train, y_train)
# 测试集构造
test_dataset = (X_test, y_test)
2.5.2 模型构建
构建一个12维的
# 模型实例化
input_size = 12
model=Linear(input_size)
2.5.3 完善Runner类
class Runner(object):
def __init__(self, model, optimizer, loss_fn, metric):
# 优化器和损失函数为None,不再关注
# 模型
self.model = model
# 评估指标
self.metric = metric
# 优化器
self.optimizer = optimizer
def train(self, dataset, reg_lambda, model_dir):
X, y = dataset
self.optimizer(self.model, X, y, reg_lambda)
# 保存模型
self.save_model(model_dir)
def evaluate(self, dataset, **kwargs):
X, y = dataset
y_pred = self.model(X)
result = self.metric(y_pred, y)
return result
def predict(self, X, **kwargs):
return self.model(X)
def save_model(self, model_dir):
if not os.path.exists(model_dir):
os.makedirs(model_dir)
params_saved_path = os.path.join(model_dir, 'params.pdtensor')
torch.save(model.params, params_saved_path)
def load_model(self, model_dir):
params_saved_path = os.path.join(model_dir, 'params.pdtensor')
self.model.params = torch.load(params_saved_path)
optimizer = optimizer_lsm
# 实例化Runner
runner = Runner(model, optimizer=optimizer, loss_fn=None, metric=mse_loss)
2.5.4 模型训练
在组装完成Runner之后,我们将开始进行模型训练、评估和测试。首先,我们先实例化Runner,然后开始进行装配训练环境,接下来就可以开始训练了,相关代码如下
# 模型保存到文件夹中
saved_dir = 'py开发'
# 启动训练
runner.train(train_dataset, reg_lambda=0, model_dir=saved_dir)
打印权重
columns_list = data.columns.to_list()
weights = runner.model.params['w'].tolist()
b = runner.model.params['b'].item()
for i in range(len(weights)):
print(columns_list[i], "weight:", weights[i])
print("b:", b)
CRIM weight: -6.7268967628479
ZN weight: 1.28081214427948
INDUS weight: -0.4696650803089142
CHAS weight: 2.235346794128418
NOX weight: -7.0105814933776855
RM weight: 9.76220417022705
AGE weight: -0.8556219339370728
DIS weight: -9.265738487243652
RAD weight: 7.973038673400879
TAX weight: -4.365403175354004
PTRATIO weight: -7.105883598327637
LSTAT weight: -13.165120124816895
b: 32.12007522583008
从输出结果看,CRIM、PTRATIO等的权重为负数,表示该镇的人均犯罪率与房价负相关,学生与教师比例越大,房价越低。RAD和CHAS等为正,表示到径向公路的可达性指数越高,房价越高;临近Charles River房价高
2.5.5 模型测试
# 加载模型权重
runner.load_model(saved_dir)
mse = runner.evaluate(test_dataset)
print('MSE:', mse.item())
MSE: 12.345974922180176
2.5.6 模型预测
runner.load_model(saved_dir)
pred = runner.predict(X_test[:1])
print("真实房价:",y_test[:1].item())
print("预测的房价:",pred.item())
真实房价: 33.099998474121094
预测的房价: 33.04654312133789
从这个结果看来真实房价和预测房价无线接近,说明结果是正确的。
实验感想:通过这次实验对线性模型有了很深彻的感悟,以后相信遇见线性模型不会再困难。