Machine Learning Pocket Reference

Chapter 19. Pipelines

Scikit-learn uses the notion of a pipeline. Using the Pipeline class, you can chain together transformers and models, and treat the whole process like a scikit-learn model. You can even insert custom logic.

Classification Pipeline

Here is an example using the tweak_titanic function inside of a pipeline:

>>> from sklearn.base import (
...     BaseEstimator,
...     TransformerMixin,
... )
>>> from sklearn.pipeline import Pipeline

>>> def tweak_titanic(df):
...     df = df.drop(
...         columns=[
...             "name",
...             "ticket",
...             "home.dest",
...             "boat",
...             "body",
...             "cabin",
...         ]
...     ).pipe(pd.get_dummies, drop_first=True)
...     return df

>>> class TitanicTransformer(
...     BaseEstimator, TransformerMixin
... ):
...     def transform(self, X):
...         # assumes X is output
...         # from reading Excel file
...         X = tweak_titanic(X)
...         X = X.drop(column="survived")
...         return X
...
...     def fit(self, X, y):
...         return self

>>> pipe = Pipeline(
...     [
...         ("titan", TitanicTransformer()),
...         ("impute", impute.IterativeImputer()),
...         (
...             "std",
...             preprocessing.StandardScaler(),
...         ),
...         ("rf", RandomForestClassifier()),
...     ]
... )

With a pipeline in hand, we can call .fit and .score on it:

>>> from sklearn.model_selection import (
...     train_test_split,
... )
>>> X_train2, X_test2, y_train2, y_test2 = train_test_split(
...     orig_df,
...     orig_df.survived,
...     test_size=0.3,
...     random_state=42,
... )

>>> pipe.fit(X_train2, y_train2)
>>> pipe.score(X_test2 ...