scikit-learn

Classical machine learning: everything except deep learning. Filed by workflow stage rather than alphabetically, because the useful question is not "what exists" but "what do I reach for now".

1 Get data Something to practise on

2 Split & tune An honest test set

3 Prepare Into numbers a model accepts

4 Compose One object, no leakage

5 Model The algorithm itself

6 Unsupervised No labels available

7 Measure Is it any good, and why

8 Utilities Glue and extension

Start here

Orientation

How to read an import

Dots are folders. Case tells you what the thing is.

Orientation

The three interfaces

Learn these and the whole library becomes predictable.

Orientation

The imports that cover most work

Learn these and you can build almost anything.

Orientation

Notes for bank work

Where the defaults are wrong for regulated data.

Orientation

Gotchas

The mistakes that do not raise an error.

1 Get data Something to practise on

sklearn.datasets

datasets

Practice data, real and synthetic.

3 worked examples

2 Split & tune An honest test set

sklearn.model_selection

model_selection

Splitting, cross-validation, hyperparameter search.

5 worked examples

3 Prepare Into numbers a model accepts

sklearn.preprocessing

preprocessing

Scaling numbers, encoding categories.

4 worked examples

sklearn.impute

impute

What to do about the blanks.

2 worked examples

sklearn.feature_extraction

feature_extraction

Text and dicts into matrices.

3 worked examples

sklearn.feature_selection

feature_selection

Which features actually earn their place.

2 worked examples

4 Compose One object, no leakage

sklearn.pipeline

pipeline

Chain steps into a single estimator.

3 worked examples

sklearn.compose

compose

Different preprocessing per column group.

4 worked examples

5 Model The algorithm itself

sklearn.linear_model

linear_model

Linear and logistic regression, and their many variants.

4 worked examples

sklearn.ensemble

ensemble

Many models combined. Where tabular accuracy lives.

4 worked examples

sklearn.tree

tree

A single decision tree. Readable, and prone to overfitting.

1 worked example

sklearn.svm

svm

Maximum-margin classifiers. Excellent on text.

1 worked example

sklearn.naive_bayes

naive_bayes

Probabilistic, absurdly fast, surprisingly hard to beat on text.

1 worked example

sklearn.neighbors

neighbors

Predict from whatever is nearby.

1 worked example

sklearn.dummy

dummy

The trivial baseline. Use it in every experiment.

1 worked example

6 Unsupervised No labels available

sklearn.cluster

cluster

Group rows without labels.

2 worked examples

sklearn.decomposition

decomposition

Fewer dimensions, most of the information.

2 worked examples

sklearn.manifold

manifold

Non-linear embedding, for looking at data.

1 worked example

sklearn.mixture

mixture

Soft clustering with probabilities.

1 worked example

7 Measure Is it any good, and why

sklearn.metrics

metrics

Is it any good — and by whose definition.

5 worked examples

sklearn.inspection

inspection

Why did it decide that.

2 worked examples

sklearn.calibration

calibration

Make predict_proba mean what it says.

2 worked examples

8 Utilities Glue and extension

sklearn.base

base

Write your own estimator that behaves like the built-ins.

2 worked examples

sklearn.utils

utils

Small helpers you will eventually want.

2 worked examples