Documentation of pmlayer

Partially monotome layer (pmlayer) is a library for neural network models. It provides several neural network layers to add monotonicity constraints on neural network models. The current library is implemented to be used in PyTorch models.

Overview:

• What is partially monotone layer?
• How to Use

How to install

pip install pmlayer

Citation

Please consider citing this paper: H. Yanagisawa, K. Miyaguchi, and T. Katsuki, “Hierarchical Lattice Layer for Partially Monotone Regression,” NeurIPS 2022.

Tutorial:

• How to use pmlayer.torch.layers.HLattice
• How to use pmlayer.torch.layers.PiecewiseLinear
• How to use custom auxiliary neural network in pmlayer.torch.layers.HLattice

API Reference:

• pmlayer.torch.layers.HLattice
• pmlayer.torch.layers.PiecewiseLinear
• pmlayer.torch.layers.PMLinear