📊 Data Science Course Hub

Feature engineering

A feature is a column the model is allowed to see. Engineering one means building it from raw columns before the split is used for fitting: a ratio, a date part, or a bin. Build it with pandas, then let the pipeline below fit encoders on the training fold only. The bank-churn SVM walkthrough does this before GridSearchCV.

PCA, without a separate lesson file

PCA redraws numeric columns along the directions that vary most. You can plot the first two components when a scatter of raw columns is unreadable. Fit it on the training fold only, inside a pipeline, after scaling. Do not treat a component as a business cause. It is a mix of the original columns.

from sklearn.decomposition import PCA
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

view = Pipeline([
    ("scale", StandardScaler()),
    ("pca", PCA(n_components=2)),
])
# Fit view on X_train only, then transform X_test with view.transform.

Save, track, and describe the model

This hub does not deploy to AWS. What you can do from the lessons: fit a pipeline, log the run, and write down what the model is allowed to decide.

import mlflow
import mlflow.sklearn

with mlflow.start_run():
    mlflow.log_param("estimator", "logistic_regression")
    mlflow.log_metric("roc_auc", float(roc_auc))
    # MLflow 2.21+ and 3.x take name=, not the older artifact_path=.
    mlflow.sklearn.log_model(pipeline, name="churn_model")

A model card next to that run should name the intended use, the training file (for example datasets/BankChurnersData.csv), the metrics, and the decisions a person must still make. Explainable AI and the Gen AI lesson are the next pages.