scikit-learn

The classical ML library for Python. Consistent API over dozens of algorithms for regression, classification, clustering, and preprocessing.

At a glance

Field	Value
Category	Classical ML library
Difficulty	Beginner
When to use	Tabular ML, prototyping, baselines, preprocessing
When not to use	Deep learning; very large out-of-core datasets
Alternatives	XGBoost, LightGBM, statsmodels

What it is

scikit-learn gives you a uniform fit / predict / transform API over almost every classical ML algorithm, plus preprocessing, model selection, and evaluation utilities. Pipelines and ColumnTransformer let you compose feature engineering and modeling into one object that you can cross-validate and pickle.

When we reach for it at Ephizen

The first model on any new tabular dataset — logistic regression, random forest, gradient boosting.
Preprocessing pipelines shared across training and serving so the feature transforms match exactly.
Quick clustering with KMeans or HDBSCAN for data exploration.
Evaluation utilities (confusion matrices, ROC curves, cross_val_score) anywhere we need them, even for PyTorch models.

Getting started

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

pipe = Pipeline([
    ("scale", StandardScaler()),
    ("clf", LogisticRegression(max_iter=1000)),
])
print(cross_val_score(pipe, X, y, cv=5).mean())

Gotchas

fit_transform on training data, transform on test data. Fitting on test data is the classic leakage bug.
For serious boosting use XGBoost or LightGBM — sklearn’s GradientBoostingClassifier is much slower.
Large datasets that don’t fit in memory need either Dask-ML, Spark MLlib, or switching strategies.
Pin scikit-learn versions when pickling models; the unpickle is fragile across versions.