Leer Automating Preprocessing with Pipelines

Automate preprocessing and feature engineering with scikit-learn pipelines to ensure consistent, reproducible machine learning results. Pipelines let you chain steps like scaling, encoding, and feature selection so every transformation always happens in the same order.

To build a pipeline in scikit-learn, define a list of steps, where each step is a tuple containing a unique step name (as a string) and a transformer object (such as StandardScaler or SelectKBest). For example:

steps = [
    ("scaler", StandardScaler()),
    ("feature_selection", SelectKBest(score_func=f_classif, k=2))
]

You then pass this list to the Pipeline object:

pipeline = Pipeline(steps)

The pipeline applies each transformer in order, passing the output of one step as the input to the next. This approach not only saves time but also reduces the risk of data leakage, making your experiments more reliable and easier to reproduce.

Using ColumnTransformer for Feature Subsets

With ColumnTransformer, you can apply different preprocessing pipelines to different subsets of features within your data. For example:

# Define column types
numeric_features = ['age', 'fare']
categorical_features = ['embarked', 'sex']

# Preprocessing for numeric features: impute missing values and scale
numeric_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='mean')),
    ('scaler', StandardScaler())
])

# Preprocessing for categorical features: impute missing values and encode
categorical_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='most_frequent')),
    ('encoder', OneHotEncoder(handle_unknown='ignore'))
])

This allows you to build a single, unified pipeline that handles both numeric and categorical data types correctly, keeping your preprocessing code organized and ensuring each transformation is applied to the intended columns.


              1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950
            
import numpy as np
import pandas as pd
import seaborn as sns
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.feature_selection import SelectKBest, f_classif

# Load the Titanic dataset from seaborn (no warnings!)
df = sns.load_dataset('titanic')

# Select features and target
features = ['age', 'fare', 'embarked', 'sex']
X = df[features]
y = df['survived']  # Target variable

# Define column types
numeric_features = ['age', 'fare']
categorical_features = ['embarked', 'sex']

# Preprocessing for numeric features: impute missing values and scale
numeric_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='mean')),
    ('scaler', StandardScaler())
])

# Preprocessing for categorical features: impute missing values and encode
categorical_transformer = Pipeline([
    ('imputer', SimpleImputer(strategy='most_frequent')),
    ('encoder', OneHotEncoder(handle_unknown='ignore'))
])

# Combine preprocessing steps
preprocessor = ColumnTransformer([
    ('num', numeric_transformer, numeric_features),
    ('cat', categorical_transformer, categorical_features)
])

# Build the full pipeline with preprocessing and feature selection
pipeline = Pipeline([
    ('preprocessing', preprocessor),
    ('feature_selection', SelectKBest(score_func=f_classif, k=3))
])

# Fit and transform the data
X_transformed = pipeline.fit_transform(X, y)

print(f"Original shape: {X.shape}")
print(f"Reduced from {X.shape[1]} features to {X_transformed.shape[1]} selected features")

Note

Integrating preprocessing into your training pipeline ensures consistent transformations and helps prevent data leakage during both training and prediction.

Was alles duidelijk?

Bedankt voor je feedback!

Sectie 1. Hoofdstuk 11

Vraag AI

Vraag wat u wilt of probeer een van de voorgestelde vragen om onze chat te starten.

Veeg om het menu te tonen

steps = [
    ("scaler", StandardScaler()),
    ("feature_selection", SelectKBest(score_func=f_classif, k=2))
]