适合Python中所有变量的模型（Scikit Learn）

Question

This was asked about a different package elsewhere, but is there a way in Scikit Learn to include all variables or all variables minus some specified number like in R?

To give an example of what I mean, say I have a regression y = x1 + x2 + x3 + x4. In RI can evaluate this regression by running:

result = lm(y ~ ., data=DF)
summary(result)

I would have to imagine there's a similar way to condense formulas in Python since writing out all the variables for larger data sets would be kind of silly.

Answer 1

is there a way in Scikit Learn to include all variables or all variables minus some specified number?

Yes, with sklearn + pandas, to fit using all variables except one, and use that one as the label, you can do simply

model.fit(df.drop('y', axis=1), df['y'])

And this would work for most sklearn models.

This would be the pandas + sklearn equivalent of R's ~ and - notation, if not using pasty .

To exclude multiple variables, you can do

df.drop(['v1', 'v2'], axis=1)

Answer 2

We may try the following workaround (let's use iris dataset and the label species as numeric and fit a linear regression model to see how to use all the independent predictors both in R and python sklearn ):

In R

summary(lm(as.numeric(Species)~., iris))[c('coefficients', 'r.squared')]

$coefficients
                Estimate Std. Error   t value     Pr(>|t|)
(Intercept)   1.18649525 0.20484104  5.792273 4.150495e-08
Sepal.Length -0.11190585 0.05764674 -1.941235 5.416918e-02
Sepal.Width  -0.04007949 0.05968881 -0.671474 5.029869e-01
Petal.Length  0.22864503 0.05685036  4.021874 9.255215e-05
Petal.Width   0.60925205 0.09445750  6.450013 1.564180e-09

$r.squared
[1] 0.9303939

In Python (sklearn with pasty)

from sklearn.datasets import load_iris
import pandas as pd
from patsy import dmatrices

iris = load_iris()
names = [f_name.replace(" ", "_").strip("_(cm)") for f_name in iris.feature_names]
iris_df = pd.DataFrame(iris.data, columns=names)
iris_df['species'] = iris.target

# pasty does not support '.' at least in windows python 2.7, so here is the workaround 
y, X = dmatrices('species ~ ' + '+'.join(iris_df.columns - ['species']),
                  iris_df, return_type="dataframe")

from sklearn.linear_model import LinearRegression
model = LinearRegression()
model.fit(X, y)

print model.score(X,y)
# 0.930422367533

print model.intercept_, model.coef_
# [ 0.19208399] [[0.22700138  0.60989412 -0.10974146 -0.04424045]]

As we can see the models learnt in R and Python with pasty are similar (the order of the coefficients are different).