⚠️ Deprecated — now a compatibility shim over mononet

As of v0.4.0a1, monotonic-nn is a thin, deprecated compatibility layer. The maintained, multi-backend implementation (PyTorch, JAX, Keras 3) lives in mononet. This package now re-exports mononet.legacy so existing airt.* imports keep working.

Breaking in 0.4.0a1: requires Keras 3 (via tensorflow>=2.16) and Python 3.11+. New code should use mononet directly. Docs: https://davorrunje.github.io/mononet/.

Constrained Monotonic Neural Networks

Summary

This package implements the Constrained Monotonic Neural Network construction described in:

Davor Runje, Sharath M. Shankaranarayana, “Constrained Monotonic Neural Networks”, in Proceedings of the 40th International Conference on Machine Learning, 2023. [PDF].

Abstract

Wider adoption of neural networks in many critical domains such as finance and healthcare is being hindered by the need to explain their predictions and to impose additional constraints on them. Monotonicity constraint is one of the most requested properties in real-world scenarios and is the focus of this paper. One of the oldest ways to construct a monotonic fully connected neural network is to constrain signs on its weights. Unfortunately, this construction does not work with popular non-saturated activation functions as it can only approximate convex functions. We show this shortcoming can be fixed by constructing two additional activation functions from a typical unsaturated monotonic activation function and employing each of them on the part of neurons. Our experiments show this approach of building monotonic neural networks has better accuracy when compared to other state-of-the-art methods, while being the simplest one in the sense of having the least number of parameters, and not requiring any modifications to the learning procedure or post-learning steps. Finally, we prove it can approximate any continuous monotone function on a compact subset of $\mathbb{R}^n$.

Status: compatibility shim

Since v0.4.0a1, this package no longer contains its own implementation. Its public API re-exports mononet.legacy, the maintained multi-backend (PyTorch / JAX / Keras 3) implementation of the same construction. The re-exported objects are the very same ones from mononet.legacy, and behaviour is numerically identical to the original TensorFlow-Keras implementation (pinned by equivalence tests in mononet).

The following continue to import, now backed by mononet and emitting a DeprecationWarning:

• airt.keras.layers.MonoDense
• airt.keras.layers.MonoDense.create_type_1 / create_type_2
• the helpers in airt._components.mono_dense_layer

New projects should depend on mononet directly.

Requirements

• Python 3.11+
• Keras 3 — installed via tensorflow>=2.16

Install

pip install --pre monotonic-nn

--pre is required while monotonic-nn and mononet are in alpha.

Usage (compatibility)

Existing code keeps working unchanged; the imports resolve to mononet.legacy. The monotonicity_indicator uses the original three-value convention — 1 for increasingly monotonic inputs, -1 for decreasingly monotonic, 0 for non-monotonic:

import keras
from keras import Sequential
from keras.layers import Input

from airt.keras.layers import MonoDense  # re-exported from mononet.legacy

model = Sequential(
    [
        Input(shape=(3,)),
        MonoDense(128, activation="elu", monotonicity_indicator=[1, 0, -1]),
        MonoDense(128, activation="elu"),
        MonoDense(1),
    ]
)

For the create_type_1 / create_type_2 builders, benchmarks, the strictly multi-backend API, and current documentation, see the mononet docs: https://davorrunje.github.io/mononet/.

Citation

If you use this library, please cite:

@inproceedings{runje2023,
  title={Constrained Monotonic Neural Networks},
  author={Davor Runje and Sharath M. Shankaranarayana},
  booktitle={Proceedings of the 40th {International Conference on Machine Learning}},
  year={2023}
}

License

This package is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0) — see LICENSE.

The underlying implementation, mononet, is available under the Apache License 2.0, which permits commercial use. If you need a permissive or commercial license, use mononet directly.