Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

老师，请问使用正规方程计算时矩阵不可逆怎么办呢？ #10

Open
eanson023 opened this issue Sep 26, 2022 · 4 comments
Open

老师，请问使用正规方程计算时矩阵不可逆怎么办呢？ #10

eanson023 opened this issue Sep 26, 2022 · 4 comments

Comments

@eanson023
Copy link

代码如下：

#! python3
# -*- encoding: utf-8 -*-
"""
Created on Mon Sep  26 08:02:10 2022

@author: eanson
"""
import pandas as pd
import torch
import math

import os

# 训练集比例
train_proportion = 0.7

#####################################################################
data_file = os.path.join('HW1 linear model', 'data1.txt')
data = pd.read_csv(data_file, header=None, names=[
                   "Room Size", "Room Amount", "Price"])
# 训练集按照比例划分
m = math.floor(len(data)*train_proportion)
X_train_pd, y_train_pd, X_cv_pd, y_cv_pd = data.iloc[:m, 0:2], data.iloc[:m,
                                                                         2], data.iloc[m+1:, 0:2], data.iloc[m+1:, 2]
# 转换为张量格式
X_train, y_train = torch.tensor(
    X_train_pd.values), torch.tensor(y_train_pd.values)
X_cv, y_cv = torch.tensor(X_cv_pd.values), torch.tensor(y_cv_pd.values)

############################################################################
# 正规方程
"""
$$
w=(X^TX)^{-1}X^Ty
$$
"""
w = X_train.t().mm(X_train).inverse().mm(X_train.t()).mm(y_train)

报错:

(d2l) yanshengdeMacBook-Pro:MachineLearning_HW_CQUT eanson$  cd /Users/eanson/Documents/ml/MachineLearning_HW_CQUT ; /usr/bin/env /Users/eanson/opt/miniconda3/envs/d2l/bin/python /Users/eanson/.vscode/extensions/ms-python.python-2021.2.633441544/pythonFiles/lib/python/debugpy/launcher 59765 -- /Users/eanson/Documents/ml/MachineLearning_HW_CQUT/linear_regression.py 
Traceback (most recent call last):
  File "/Users/eanson/Documents/ml/MachineLearning_HW_CQUT/linear_regression.py", line 37, in <module>
    w = X_train.t().mm(X_train).inverse().mm(X_train.t()).mm(y_train)
RuntimeError: "linalg_inv_out_cpu" not implemented for 'Long'

具体就是调用Inverse函数时，前面的矩阵不可逆😄
@mousecpn
Copy link
Owner

代码如下：

#! python3
# -*- encoding: utf-8 -*-
"""
Created on Mon Sep  26 08:02:10 2022

@author: eanson
"""
import pandas as pd
import torch
import math

import os

# 训练集比例
train_proportion = 0.7

#####################################################################
data_file = os.path.join('HW1 linear model', 'data1.txt')
data = pd.read_csv(data_file, header=None, names=[
                   "Room Size", "Room Amount", "Price"])
# 训练集按照比例划分
m = math.floor(len(data)*train_proportion)
X_train_pd, y_train_pd, X_cv_pd, y_cv_pd = data.iloc[:m, 0:2], data.iloc[:m,
                                                                         2], data.iloc[m+1:, 0:2], data.iloc[m+1:, 2]
# 转换为张量格式
X_train, y_train = torch.tensor(
    X_train_pd.values), torch.tensor(y_train_pd.values)
X_cv, y_cv = torch.tensor(X_cv_pd.values), torch.tensor(y_cv_pd.values)

############################################################################
# 正规方程
"""
$$
w=(X^TX)^{-1}X^Ty
$$
"""
w = X_train.t().mm(X_train).inverse().mm(X_train.t()).mm(y_train)

报错:

(d2l) yanshengdeMacBook-Pro:MachineLearning_HW_CQUT eanson$  cd /Users/eanson/Documents/ml/MachineLearning_HW_CQUT ; /usr/bin/env /Users/eanson/opt/miniconda3/envs/d2l/bin/python /Users/eanson/.vscode/extensions/ms-python.python-2021.2.633441544/pythonFiles/lib/python/debugpy/launcher 59765 -- /Users/eanson/Documents/ml/MachineLearning_HW_CQUT/linear_regression.py 
Traceback (most recent call last):
  File "/Users/eanson/Documents/ml/MachineLearning_HW_CQUT/linear_regression.py", line 37, in <module>
    w = X_train.t().mm(X_train).inverse().mm(X_train.t()).mm(y_train)
RuntimeError: "linalg_inv_out_cpu" not implemented for 'Long'

具体就是调用Inverse函数时，前面的矩阵不可逆😄

看报错好像并不是矩阵不可逆，而是inverse这个函数不能对Long型变量操作。
可以先把数据转成float型变量

@eanson023
Copy link
Author

还真是，我改了之后能跑起来了,谢谢老师

@eanson023
Copy link
Author

eanson023 commented Sep 26, 2022 •
edited
Loading

老师，我这个loss虽然下降了，但是太大了，您能帮我看下具体哪出了问题吗?

输出：

iter:9991,loss:2224352768.0
iter:9992,loss:2224352768.0
iter:9993,loss:2224352768.0
iter:9994,loss:2224352768.0
iter:9995,loss:2224352768.0
iter:9996,loss:2224352768.0
iter:9997,loss:2224352768.0
iter:9998,loss:2224352768.0
iter:9999,loss:2224352768.0
iter:10000,loss:2224352768.0

loss图:
image

代码如下:

#! python3
# -*- encoding: utf-8 -*-
"""
Created on Mon Sep  26 08:02:10 2022

@author: eanson
"""
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import torch
import math

import os

# 训练集比例
train_proportion = 0.7
figure_idx = 1

############################################################################

# 读取数据集
data_file = os.path.join('HW1 linear model', 'data1.txt')
data = pd.read_csv(data_file, header=None, names=[
                   "Room Size", "Room Amount", "Price"])
# 训练集按照比例划分
m = math.floor(len(data)*train_proportion)
X_train_pd, y_train_pd, X_cv_pd, y_cv_pd = data.iloc[:m, 0:2], data.iloc[:m,
                                                                         2], data.iloc[m+1:, 0:2], data.iloc[m+1:, 2]
# 转换为张量格式 引入特征 x0=1 （方便之后向量化）
X_train, y_train = torch.cat([torch.ones(m, 1), torch.tensor(
    X_train_pd.values, dtype=torch.float)], dim=1), torch.tensor(y_train_pd.values, dtype=torch.float).reshape(-1, 1)
X_cv, y_cv = torch.cat([torch.ones(len(X_cv_pd), 1), torch.tensor(
    X_cv_pd.values, dtype=torch.float)], dim=1), torch.tensor(y_cv_pd.values, dtype=torch.float).reshape(-1, 1)

# 可视化
def visualize(X, y, w, title_suffix):
    global figure_idx
    # 计算验证集预测值
    y_pred = torch.mm(X, w)
    x_points = X[:, 1].numpy()
    y_points = y.detach().numpy()
    y_pred_points = y_pred.detach().numpy()

    plt.figure(figure_idx)
    figure_idx += 1
    plt.title("house price prediction "+title_suffix)
    plt.xlabel("room size(m²)")
    plt.ylabel("house price($)")
    plt.plot(x_points, y_points, 'o', c='b')
    plt.plot(x_points, y_pred_points, 'o', c='r')
    plt.show()


############################################################################
# 梯度下降法
# feature scaling 将房间大小进行特征缩放
"""
$$
x_{n} = \frac{x_{n}-u_{n}}{s_{n}} （其中u_{n}是平均值，s_{n}是标准差）
$$
"""
X_train[:, 1] = (X_train[:, 1]-X_train[:, 1].mean())/X_train[:, 1].std()

# 迭代次数
iterations = 10000
# 学习率
lr = 0.1
# 初始化theta
w = torch.rand((X_train.size(1), 1))  # size: feature_size+1 x 1


def loss_function(y_pred, y):
    m = len(y_pred)
    return 1/(2*m)*torch.sum((y_pred-y)**2)


log = []
for i in range(iterations):
    # 假设函数
    hypothesis = torch.mm(X_train, w)
    # 向量化
    grad = lr*torch.mean((hypothesis-y_train)*X_train, dim=0).reshape(-1, 1)
    # 梯度下降
    w = w - grad
    loss = loss_function(hypothesis, y_train)
    print(f'iter:{i+1},loss:{loss}')
    log.append([i+1, loss])

# 画下loss
log = np.array(log)
plt.figure(figure_idx)
figure_idx += 1
plt.plot(log[:, 0], log[:, 1])
plt.show()

visualize(X_cv, y_cv, w, 'on cross validation data with gradient descent')

@mousecpn
Copy link
Owner

老师，我这个loss虽然下降了，但是太大了，您能帮我看下具体哪出了问题吗?

输出：

iter:9991,loss:2224352768.0
iter:9992,loss:2224352768.0
iter:9993,loss:2224352768.0
iter:9994,loss:2224352768.0
iter:9995,loss:2224352768.0
iter:9996,loss:2224352768.0
iter:9997,loss:2224352768.0
iter:9998,loss:2224352768.0
iter:9999,loss:2224352768.0
iter:10000,loss:2224352768.0

loss图: image

代码如下:

#! python3
# -*- encoding: utf-8 -*-
"""
Created on Mon Sep  26 08:02:10 2022

@author: eanson
"""
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import torch
import math

import os

# 训练集比例
train_proportion = 0.7
figure_idx = 1

############################################################################

# 读取数据集
data_file = os.path.join('HW1 linear model', 'data1.txt')
data = pd.read_csv(data_file, header=None, names=[
                   "Room Size", "Room Amount", "Price"])
# 训练集按照比例划分
m = math.floor(len(data)*train_proportion)
X_train_pd, y_train_pd, X_cv_pd, y_cv_pd = data.iloc[:m, 0:2], data.iloc[:m,
                                                                         2], data.iloc[m+1:, 0:2], data.iloc[m+1:, 2]
# 转换为张量格式 引入特征 x0=1 （方便之后向量化）
X_train, y_train = torch.cat([torch.ones(m, 1), torch.tensor(
    X_train_pd.values, dtype=torch.float)], dim=1), torch.tensor(y_train_pd.values, dtype=torch.float).reshape(-1, 1)
X_cv, y_cv = torch.cat([torch.ones(len(X_cv_pd), 1), torch.tensor(
    X_cv_pd.values, dtype=torch.float)], dim=1), torch.tensor(y_cv_pd.values, dtype=torch.float).reshape(-1, 1)

# 可视化
def visualize(X, y, w, title_suffix):
    global figure_idx
    # 计算验证集预测值
    y_pred = torch.mm(X, w)
    x_points = X[:, 1].numpy()
    y_points = y.detach().numpy()
    y_pred_points = y_pred.detach().numpy()

    plt.figure(figure_idx)
    figure_idx += 1
    plt.title("house price prediction "+title_suffix)
    plt.xlabel("room size(m²)")
    plt.ylabel("house price($)")
    plt.plot(x_points, y_points, 'o', c='b')
    plt.plot(x_points, y_pred_points, 'o', c='r')
    plt.show()


############################################################################
# 梯度下降法
# feature scaling 将房间大小进行特征缩放
"""
$$
x_{n} = \frac{x_{n}-u_{n}}{s_{n}} （其中u_{n}是平均值，s_{n}是标准差）
$$
"""
X_train[:, 1] = (X_train[:, 1]-X_train[:, 1].mean())/X_train[:, 1].std()

# 迭代次数
iterations = 10000
# 学习率
lr = 0.1
# 初始化theta
w = torch.rand((X_train.size(1), 1))  # size: feature_size+1 x 1


def loss_function(y_pred, y):
    m = len(y_pred)
    return 1/(2*m)*torch.sum((y_pred-y)**2)


log = []
for i in range(iterations):
    # 假设函数
    hypothesis = torch.mm(X_train, w)
    # 向量化
    grad = lr*torch.mean((hypothesis-y_train)*X_train, dim=0).reshape(-1, 1)
    # 梯度下降
    w = w - grad
    loss = loss_function(hypothesis, y_train)
    print(f'iter:{i+1},loss:{loss}')
    log.append([i+1, loss])

# 画下loss
log = np.array(log)
plt.figure(figure_idx)
figure_idx += 1
plt.plot(log[:, 0], log[:, 1])
plt.show()

visualize(X_cv, y_cv, w, 'on cross validation data with gradient descent')

没做归一化

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@mousecpn @eanson023