Building and Deploying Machine Learning Pipelines with scikit-learn | by Noor Fatima | Jun, 2024

On this article, we’ll stroll by means of the method of constructing and deploying machine studying pipelines utilizing the Pipeline class from scikit-learn. We are going to use a dataset from the Titanic competitors as an example the method.

A machine studying pipeline in scikit-learn is a solution to streamline a collection of information processing and modeling steps. Pipelines assist be sure that the identical transformations are utilized throughout each coaching and testing, stopping information leakage and making your workflow cleaner and extra reproducible.

We are going to use the Titanic dataset, which accommodates details about passengers and whether or not they survived the Titanic catastrophe. The objective is to construct a mannequin that predicts survival primarily based on passenger attributes.

import pandas as pd
# Load the dataset
df = pd.read_csv('practice.csv')
print(df.head())

We drop columns that gained’t be helpful for prediction.

df.drop(columns=['PassengerId', 'Name', 'Ticket', 'Cabin'], inplace=True)

Cut up the info into coaching and testing units.

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(
df.drop(columns=['Survived']), 
df['Survived'], 
test_size=0.2, 
random_state=42
)

Imputation Transformer

Deal with lacking values.

from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
trf1 = ColumnTransformer([
('impute_age', SimpleImputer(), [2]),  # Impute Age
('impute_embarked', SimpleImputer(technique='most_frequent'), [6])  # Impute Embarked
], the rest='passthrough')

One-Scorching Encoding

Convert categorical variables into numeric.

from sklearn.preprocessing import OneHotEncoder
trf2 = ColumnTransformer([
('ohe_sex_embarked', OneHotEncoder(sparse=False, handle_unknown='ignore'), [1, 6])  # One-Scorching Encode Intercourse and Embarked
], the rest='passthrough')

Scaling

Scale the options to a given vary.

from sklearn.preprocessing import MinMaxScaler
trf3 = ColumnTransformer([
('scale', MinMaxScaler(), slice(0, 10))  # Scale all features
])

Function Choice

Choose an important options.

from sklearn.feature_selection import SelectKBest, chi2
trf4 = SelectKBest(score_func=chi2, ok=8)

Use a call tree classifier.

from sklearn.tree import DecisionTreeClassifier
trf5 = DecisionTreeClassifier()

Mix all transformations and the mannequin right into a single pipeline.

from sklearn.pipeline import Pipeline
pipe = Pipeline([
('trf1', trf1),
('trf2', trf2),
('trf3', trf3),
('trf4', trf4),
('trf5', trf5)
])
# Prepare the pipeline
pipe.match(X_train, y_train)

Consider the mannequin on the take a look at information.

from sklearn.metrics import accuracy_score
y_pred = pipe.predict(X_test)
print(accuracy_score(y_test, y_pred))

Use cross-validation to test the mannequin’s robustness.

from sklearn.model_selection import cross_val_score
print(cross_val_score(pipe, X_train, y_train, cv=5, scoring='accuracy').imply())

Use grid search to seek out the very best hyperparameters.

from sklearn.model_selection import GridSearchCV
params = {
'trf5__max_depth': [1, 2, 3, 4, 5, None]
}
grid = GridSearchCV(pipe, params, cv=5, scoring='accuracy')
grid.match(X_train, y_train)
print(grid.best_score_)
print(grid.best_params_)

Export the skilled pipeline to a file for later use.

import pickle
pickle.dump(pipe, open('pipe.pkl', 'wb'))

Load the pipeline and use it for predictions.

pipe = pickle.load(open('pipe.pkl', 'rb'))
# Instance consumer enter
test_input = np.array([2, 'male', 31.0, 0, 0, 10.5, 'S'], dtype=object).reshape(1, 7)
print(pipe.predict(test_input))

Pipelines in scikit-learn present a strong solution to handle the whole machine studying workflow, from preprocessing to mannequin coaching and analysis. By following this information, you’ll be able to construct strong and reproducible pipelines on your personal machine studying initiatives.