Monotonicnetworks

Latest version: v1.5.1

Safety actively analyzes 681812 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

1.5.0

Implemented various layers for convenience.
In addition to `direct_norm` now we have native Lipschitz and Monotonic layers:

The `LipschitzLinear` class is a linear layer with a Lipschitz constraint on its weights.
The `MonotonicLayer` class is a linear layer with a Lipschitz constraint on its weights and monotonicity constraints that can be specified for each input dimension, or for each input-output pair.
The `MonotonicWrapper` class is a wrapper around a module with a Lipschitz constant. It adds a term to the output of the module which enforces monotonicity constraints given by monotone_constraints. The class returns a module that is monotonic and Lipschitz with constant lipschitz_const.
The `SigmaNet` class is a deprecated class that is equivalent to the MonotonicWrapper class.
The `RMSNorm` class is a class that implements the RMSNorm normalization layer. It can help when training
a model with many Lipschitz/MontonicLayers.
Updated the main README.md with more details and examples.

Updated plots and added code for the toy figures in the paper.

*Breaking Changes!*
`monotonenorm` has been renamed to `monotonicnetworks`.
New PyPI name following the same change. The old package name will still exist but will be marked as deprecated.

1.0.0

The first release of monotonenorm. Compatible with PyTorch 1.12.1.

Links

Releases

Has known vulnerabilities

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.