在python中查找逻辑回归的系数
Posted
技术标签:
【中文标题】在python中查找逻辑回归的系数【英文标题】：Finding coefficients for logistic regression in python 【发布时间】：2020-01-15 09:07:56 【问题描述】：我正在研究一个分类问题，需要逻辑回归方程的系数。我可以在 R 中找到系数，但我需要在 python 中提交项目。我在 python 中找不到学习逻辑回归系数的代码。 python中如何获取系数值？
【问题讨论】：
clf.coef_, clf.intercept_ 分别是权重和偏差。检查 coef_ 的大小，它是 coef_ndarray of shape (1, n_features) or (n_classes, n_features)。
【参考方案1】：
提供更多细节并展示如何替换 pytorch 模型的最后一层：
#%%
"""
Get the weights & biases to set them to a nn.Linear layer in pytorch
"""
import numpy as np
import torch
from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression
from torch import nn
X, y = load_iris(return_X_y=True)
print(f'X.shape=')
print(f'y.shape=')
Din: int = X.shape[1]
total_data_set_size: int = X.shape[0]
assert y.shape[0] == total_data_set_size
clf = LogisticRegression(random_state=0).fit(X, y)
out = clf.predict(X[:2, :])
# print(f'out=')
out = clf.predict_proba(X[:2, :])
print(f'out=')
clf.score(X, y)
# - coef_ndarray of shape (1, n_features) or (n_classes, n_features)
print(f'clf.coef_.shape=')
print(f'clf.intercept_.shape=')
assert (clf.coef_.shape[1] == Din)
Dout: int = clf.coef_.shape[0]
print(f'Dout= which is the number of classes too in classification')
assert (Dout == clf.intercept_.shape[0])
print()
num_classes: int = Dout
mdl = nn.Linear(in_features=Din, out_features=num_classes)
mdl.weight = torch.nn.Parameter(torch.from_numpy(clf.coef_))
mdl.bias = torch.nn.Parameter(torch.from_numpy(clf.intercept_))
out2 = torch.softmax(mdl(torch.from_numpy(X[:2, :])), dim=1)
print(f'out2=')
assert np.isclose(out2.detach().cpu().numpy(), out).all()
# -
# module: nn.Module = getattr(base_model, layer_to_replace)
# num_classes: int = clf.coef_[0] # out_features=Dout
# num_features: int = clf.coef_[1] # in_features
# assert module.weight.Size() == torch.Size([num_features, num_classes])
# assert module.bias.Size() == torch.Size([num_classes])
# module.weight = torch.nn.Parameter(torch.from_numpy(clf.coef_))
# module.bias = torch.nn.Parameter(torch.from_numpy(clf.intercept_))
【讨论】：
【参考方案2】：最后一个答案稍微更正：
pd.DataFrame(zip(X_train.columns, np.transpose(clf.coef_.tolist()[0])), columns=['features', 'coef'])
【讨论】：
【参考方案3】：假设您的 X 是 Pandas DataFrame 并且 clf 是您的逻辑回归模型，您可以使用这行代码获取功能的名称及其值：
pd.DataFrame(zip(X_train.columns, np.transpose(clf.coef_)), columns=['features', 'coef'])
【讨论】：
【参考方案4】：statsmodels 库将为您提供系数结果的细分以及相关的 p 值以确定它们的重要性。
使用 x1 和 y1 变量的示例：
x1_train, x1_test, y1_train, y1_test = train_test_split(x1, y1, random_state=0)
logreg = LogisticRegression().fit(x1_train,y1_train)
logreg
print("Training set score: :.3f".format(logreg.score(x1_train,y1_train)))
print("Test set score: :.3f".format(logreg.score(x1_test,y1_test)))
import statsmodels.api as sm
logit_model=sm.Logit(y1,x1)
result=logit_model.fit()
print(result.summary())
示例结果：
Optimization terminated successfully.
Current function value: 0.596755
Iterations 7
Logit Regression Results
==============================================================================
Dep. Variable: IsCanceled No. Observations: 20000
Model: Logit Df Residuals: 19996
Method: MLE Df Model: 3
Date: Sat, 17 Aug 2019 Pseudo R-squ.: 0.1391
Time: 23:58:55 Log-Likelihood: -11935.
converged: True LL-Null: -13863.
LLR p-value: 0.000
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
const -2.1417 0.050 -43.216 0.000 -2.239 -2.045
x1 0.0055 0.000 32.013 0.000 0.005 0.006
x2 0.0236 0.001 36.465 0.000 0.022 0.025
x3 2.1137 0.104 20.400 0.000 1.911 2.317
==============================================================================
【讨论】：
【参考方案5】：路飞，请记住始终分享您的代码和尝试，以便我们了解您的尝试并为您提供帮助。无论如何，我认为您正在寻找这个：
import numpy as np
from sklearn.linear_model import LogisticRegression
X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]]) #Your x values, for a 2 variable model.
#y = 1 * x_0 + 2 * x_1 + 3 #This is the "true" model
y = np.dot(X, np.array([1, 2])) + 3 #Generating the true y-values
reg = LogisticRegression().fit(X, y) #Fitting the model given your X and y values.
reg.coef_ #Prints an array of all regressor values (b1 and b2, or as many bs as your model has)
reg.intercept_ #Prints value for intercept/b0
reg.predict(np.array([[3, 5]])) #Predicts an array of y-values with the fitted model given the inputs
【讨论】：
【参考方案6】：sklearn.linear_model.LogisticRegression 适合您。 看这个例子：
from sklearn.linear_model import LogisticRegression
from sklearn.datasets import load_iris
X, y = load_iris(return_X_y=True)
clf = LogisticRegression(random_state=0).fit(X, y)
print(clf.coef_, clf.intercept_)
【讨论】：
【参考方案7】：看看statsmodels library's Logit model。
你会这样使用它：
from statsmodels.discrete.discrete_model import Logit
from statsmodels.tools import add_constant
x = [...] # Obesrvations
y = [...] # Response variable
x = add_constant(x)
print(Logit(y, x).fit().summary())
【讨论】：
以上是关于在python中查找逻辑回归的系数的主要内容，如果未能解决你的问题，请参考以下文章
python：如何在sklearn中使用逻辑回归系数构建决策边界