Scikit-Learn Pipeline

Introduction

Pipeline chains multiple transformers with a final estimator for streamlined ML workflows.

Creating Pipelines

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

# Manual pipeline creation
pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', LogisticRegression())
])

# Fit on data
X = [[1, 2], [3, 4], [5, 6]]
y = [0, 1, 0]
pipeline.fit(X, y)
prediction = pipeline.predict([[2, 3]])

make_pipeline

from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC

# Shorthand syntax - auto-names steps
model = make_pipeline(StandardScaler(), SVC(kernel='linear'))

# Works identically
model.fit(X, y)
result = model.predict([[0, 1]])

Pipeline with Preprocessing

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
import pandas as pd

# Different transformations for different columns
preprocessor = ColumnTransformer([
    ('num', StandardScaler(), ['age', 'income']),
    ('cat', OneHotEncoder(), ['city', 'occupation'])
])

pipeline = Pipeline([
    ('preprocess', preprocessor),
    ('classifier', LogisticRegression())
])

Accessing Pipeline Steps

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LinearRegression

pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('regressor', LinearRegression())
])
pipeline.fit([[1], [2], [3]], [1, 2, 3])

# Access individual steps
scaler = pipeline.named_steps['scaler']
regressor = pipeline.named_steps['regressor']

# Get feature names after preprocessing
print(regressor.coef_)

Practice Problems

1. Create pipeline with scaler and classifier
2. Use make_pipeline for quick setup
3. Add feature selection to pipeline
4. Access individual step parameters
5. Use ColumnTransformer for mixed data types