Registry / ai-ml / pytorch-tabnet

pytorch-tabnet

library4.1.0pypypiunverified

PyTorch implementation of TabNet (Google's attention-based tabular network). Current version 4.1.0, with semi-annual releases. Supports classification, regression, and unsupervised pre-training.

ai-ml

pip install pytorch-tabnet

github ↗package ↗docs ↗

INSTALL

IMPORT

SIG · PYTORCH-TABNET

pytorch-tabnet

ai-mlpythonv4.1.0

Install

74.8s avg

Import

—

Disk

5018MB

Pass rate

4/ 10

Env Coverage4 / 10

glibc

3.9–3.13

musl

3.9–3.13

Install & Compatibility

Where this runs

tested against v4.1.0 · pip install

no network on importno background threads

Install × environment matrix

Each cell = how many times install + import succeeded across repeated harness runs. Partial = flaky.
glibc = Debian/Ubuntu slim · musl = Alpine Linux

musl

glibc

py 3.10

✕ build_error

✓ 88.2s

py 3.11

✕ build_error

✓ 77.7s

py 3.12

✕ build_error

✓ 68.8s

py 3.13

✕ build_error

✓ 64.7s

py 3.9

✕ build_error

✕ timeout

5018MB installed

● package 5018MB

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

TabNetClassifier

✓ from pytorch_tabnet import TabNetClassifier

✗ from pytorch_tabnet.tab_model import TabNetClassifier

Minimal example of fitting a TabNetClassifier on random data.

from pytorch_tabnet.tab_model import TabNetClassifier import numpy as np X_train = np.random.rand(100, 10) y_train = np.random.randint(0, 2, 100) clf = TabNetClassifier(device_name='cpu') clf.fit(X_train, y_train, max_epochs=10) print(clf.predict(X_train))

Debug

Known issues

breakingIn v4.0, the unsupervised pretraining loss was changed to match the original paper. Models trained with pretraining in v3.x cannot be directly resumed or fine-tuned in v4.x without retraining.

fix

Retrain any models that used unsupervised pretraining after upgrading to v4.x.

affects: v4.0+

breakingThe default metric for regression changed from 'mse' to 'rmse' in v3.0.0. If you relied on default metric behavior, your training/evaluation results may differ.

fix

Explicitly set the `eval_metric` parameter to 'mse' to retain old behavior.

affects: v3.0.0+

gotchaWhen using categorical features, `cat_idxs` and `cat_dims` must be consistent. An error is raised only in v4.0+ if they are incoherent.

fix

Ensure `len(cat_idxs) == len(cat_dims)` and that each index corresponds to a valid column.

affects: >=4.0

gotchaSaving and loading models: use `save_model` and `load_model` methods. Directly pickling the model object is not supported and may break.

fix

Use clf.save_model('model.pt') and clf.load_model('model.pt').

affects: all

Upgrade

Version history

4.1.0latest on PyPI · released Jul 23, 2023

Audit

Dependencies

No dependency data recorded yet.

Agent activity

13 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/dreamquark-ai/tabnet ↗

packagehttps://pypi.org/project/pytorch-tabnet/ ↗

docshttps://github.com/dreamquark-ai/tabnet ↗