Home

torml Wiki

torml is a scikit-learn-style machine learning library implemented from scratch with PyTorch (torch.Tensor, torch.linalg) as its numerical backend, instead of NumPy/SciPy.

• Repository: https://github.com/ms-kumar/torml
• License: MIT
• Release notes: RELEASES.md
• Contributing: CONTRIBUTING.md
• Security: SECURITY.md — report vulnerabilities privately

Install

pip install -e .
pip install -e ".[test]"   # pytest
pip install -e ".[doc]"    # sphinx docs

With uv:

uv sync --extra dev --extra test
uv run pytest
uv run pre-commit run --all-files

Quickstart

import torch
from torml.linear_model import LinearRegression

X = torch.randn(50, 3)
y = X @ torch.tensor([1.0, 2.0, -1.0]) + 0.1

model = LinearRegression().fit(X, y)
model.predict(X[:5])

Modules

Module	Status	Contents
torml.base	done (Phase 1)	BaseEstimator, clone, mixins, check_estimator
torml.utils	done (Phase 1)	validation, random state, tags, masks
torml.metrics	done (Phase 2)	accuracy_score, mean_squared_error, r2_score
torml.linear_model	done (Phase 2)	LinearRegression, LogisticRegression
torml.model_selection	done (Phase 2)	train_test_split, KFold, cross_val_score
torml.preprocessing	done (Phase 2)	StandardScaler, MinMaxScaler, LabelEncoder, OneHotEncoder
torml.neighbors	done	KNeighborsClassifier, KNeighborsRegressor
torml.naive_bayes	done	GaussianNB
torml.cluster	done	KMeans, DBSCAN
torml.tree	done	DecisionTreeClassifier, DecisionTreeRegressor
torml.decomposition	done	PCA
torml.ensemble	done	VotingClassifier/Regressor, BaggingClassifier/Regressor, RandomForestClassifier/Regressor
torml.svm	done	LinearSVC, LinearSVR
torml.pipelines	done	Pipeline, FeatureUnion, ColumnTransformer
torml.manifold	done	MDS
torml.gaussian_process	done	GaussianProcessRegressor
torml.mixture	done	GaussianMixture
torml.multiclass	done	OneVsRestClassifier
torml.semi_supervised	done	LabelPropagation
torml.covariance	done	EmpiricalCovariance
torml.cross_decomposition	done	PLSRegression
torml.feature_extraction	done	DictVectorizer
torml.feature_selection	done	SelectKBest, f_classif
torml.random_projection	done	GaussianRandomProjection
torml.discriminant_analysis	done	LinearDiscriminantAnalysis
torml.multivariate	done	MultiOutputRegressor, MultiOutputClassifier

285 tests, CI + Tests green on main. Release v0.1.0.

Branches

• main — integration (CI green, v0.1.0 released).
• Phase0 — repo scaffold milestone (frozen).
• Phase1 — base + utils milestone (frozen).
• Phase2 — first vertical slice milestone (frozen).
• Phase3 — full backlog milestone (merged to main).

Each phase builds on the previous one: Phase0 ⊂ Phase1 ⊂ Phase2 ⊂ Phase3.

Each phase builds on the previous one: Phase0 ⊂ Phase1 ⊂ Phase2 ⊂ Phase3.

Pages

• Estimator conventions — fit/predict/transform/score, hyperparameters vs learned attributes.
• Validation utilities — check_array, check_X_y, check_is_fitted, check_random_state.