时间戳

首页 > 解决方案 > Use existing coefficient and intercept in Linear regression

python - Use existing coefficient and intercept in Linear regression

问题描述

I'm using scikit-learn module for Linear Regression. My model runs every day, now i store the model's variable (coef_ & intercept_) to file, so that it can be used again when my model run.

Let's suppose, on a history of one year, i'm running the model daily. On 25th of November, i saved the model's coef_ & intercept_ in a file, so again i restart my program, and it will start from 25th of November, and run till the last

So when i compare the predictions for 26th November before and after the restarting, predictions are different. So i just thought of using the coef_ & intercept_ before restarting, so that after restart, it should predict the same for 26th November.

To do this, i just overwrite the coef_ & intercept_

from sklearn import linear_model

model = linear_model.LinearRegression()
model.coef_ = coef_stored
model.intercept_ = intercept_stored

model.fit(X, y)
model.predict(x)

I want my predictions for 26th to be same, before and after the restart. Using above code i was not able to achieve it.

标签： pythonpython-3.xscikit-learndata-sciencelinear-regression

解决方案

不建议更改未经训练的模型的属性，但按照 Desertnaut 的评论，您可以按照How to instantiate a Scikit-Learn linear model with known coefficients without fit it中所示进行操作。

但是，如果您调用该fit方法，则系数和截距将被覆盖。

from sklearn.linear_model import LinearRegression
import numpy as np
np.random.seed(0)

my_intercepts = np.ones(2)
my_coefficients = np.random.randn(2, 3)

new_model = LinearRegression()
new_model.intercept_ = my_intercepts
new_model.coef_ = my_coefficients

print(new_model.coef_)
#[[ 1.76405235  0.40015721  0.97873798]
# [ 2.2408932   1.86755799 -0.97727788]]


new_model.predict(np.random.randn(5, 3))
#array([[ 2.51441481,  2.94725181],
#       [ 3.20531004,  0.76788778],
#       [ 2.82562532,  2.49886169],
#       [ 1.98568931,  4.73850448],
#       [-1.28821286,  2.60145844]])

你说：

所以我只是想在重新启动之前使用 coef_ 和 intercept_，这样在重新启动后，它应该预测 11 月 26 日的情况相同。

如果在确定使用相同的数据和模型系数时没有得到相同的结果，则说明有问题。如果您在程序的前后状态之间进行sklearn升级，则不同版本之间可能会有细微的差异。sklearn

推荐阅读