linear-model-feature-importance-calculation.md

Linear model feature importance calculation

For linear models, trained coefficients can be used as feature importance values:

from sklearn import datasets, linear_model, model_selection, metrics

X, y = datasets.load_diabetes(return_X_y=True)
X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y)

model = linear_model.LinearRegression()
model.fit(X_train, y_train)

fi = model.coef_

• from sklearn import - import module from lib:scikit-learn
• datasets.load_diabetes - loads sample diabetes database
• model_selection.train_test_split - splits given X and y datasets to test (25% of values by default) and train (75% of values by default) subsets
• .fit( - train model with a given features and target variable dataset
• .coef_ - returns list of coefficients of a trained model

group: feature-importance

Example:

from sklearn import datasets, linear_model, model_selection, metrics

X, y = datasets.load_diabetes(return_X_y=True)
X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y)

model = linear_model.LinearRegression()
model.fit(X_train, y_train)

fi = model.coef_
print(fi)

[   7.4644251  -212.29645468  484.41681905  275.862333   -938.22675656
  587.40072837  114.72619725  120.06905393  872.88971664   45.45492861]