Engression
Engression is a neural network-based distributional regression method proposed in the paper "Engression: Extrapolation through the Lens of Distributional Regression?" by Xinwei Shen and Nicolai Meinshausen (2023). This repository contains the software implementations of engression in both R and Python.
Consider targets $Y\in\mathbb{R}^k$ and predictors $X\in\mathbb{R}^d$; both variables can be univariate or multivariate, continuous or discrete. Engression can be used to
- estimate the conditional mean $\mathbb{E}[Y|X=x]$ (as in least-squares regression),
- estimate the conditional quantiles of $Y$ given $X=x$ (as in quantile regression), and
- sample from the fitted conditional distribution of $Y$ given $X=x$ (as a generative model).
The results in the paper show the advantages of engression over existing regression approaches in terms of extrapolation.
Installation
The latest release of the Python package can be installed through pip:
pip install engression
The development version can be installed from github:
pip install -e "git+https://github.com/xwshen51/engression#egg=engression&subdirectory=engression-python"
Usage Example
Python
Below is one simple demonstration. See this tutorial for more details on simulated data and this tutorial for a real data example.
from engression import engression
from engression.data.simulator import preanm_simulator
## Simulate data
x, y = preanm_simulator("square", n=10000, x_lower=0, x_upper=2, noise_std=1, train=True, device="cpu")
x_eval, y_eval_med, y_eval_mean = preanm_simulator("square", n=1000, x_lower=0, x_upper=4, noise_std=1, train=False, device="cpu")
## Fit an engression model
engressor = engression(x, y, lr=0.01, num_epochs=500, batch_size=1000, device="cpu")
## Summarize model information
engressor.summary()
## Evaluation
print("L2 loss:", engressor.eval_loss(x_eval, y_eval_mean, loss_type="l2"))
print("correlation between predicted and true means:", engressor.eval_loss(x_eval, y_eval_mean, loss_type="cor"))
## Predictions
y_pred_mean = engressor.predict(x_eval, target="mean") ## for the conditional mean
y_pred_med = engressor.predict(x_eval, target="median") ## for the conditional median
y_pred_quant = engressor.predict(x_eval, target=[0.025, 0.5, 0.975]) ## for the conditional 2.5% and 97.5% quantiles
Validation data can be passed to choose the number of training iterations: engression(x, y, x_val=x_val, y_val=y_val) computes the energy loss on them about every 50 iterations and after the last epoch, and keeps the parameters with the lowest loss; engressor.summary() shows the epoch they come from.
Contact information
If you meet any problems with the code, please submit an issue or contact Xinwei Shen.
Citation
If you would refer to or extend our work, please cite the following paper:
@article{10.1093/jrsssb/qkae108,
author = {Shen, Xinwei and Meinshausen, Nicolai},
title = {Engression: extrapolation through the lens of distributional regression},
journal = {Journal of the Royal Statistical Society Series B: Statistical Methodology},
pages = {qkae108},
year = {2024},
month = {11},
issn = {1369-7412},
doi = {10.1093/jrsssb/qkae108},
url = {https://doi.org/10.1093/jrsssb/qkae108},
eprint = {https://academic.oup.com/jrsssb/advance-article-pdf/doi/10.1093/jrsssb/qkae108/60827977/qkae108.pdf},
}
Metadata
Release files for engression 1.0.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| engression-1.0.0.tar.gz | 26.9 kB | Without provenance | Details for engression-1.0.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| engression-1.0.0-py3-none-any.whl | Python 3 | none | any | Details for engression-1.0.0-py3-none-any.whl |
Total release size: 47.2 kB
Release files / engression-1.0.0.tar.gz
| Download URL | engression-1.0.0.tar.gz |
|---|---|
| Size | 26.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
6b6891d847e2df3886d8def22bdbf72f825de145471832a36ec603ca6e0af9bd
|
|
BLAKE2b-256 checksum How to use checksums |
774c41815b5e4acb42aabf1df7b7d4e401b35af797fe9cc281fe3bd535709797
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.2 CPython/3.11.2
|
Release files / engression-1.0.0-py3-none-any.whl
| Download URL | engression-1.0.0-py3-none-any.whl |
|---|---|
| Size | 20.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
68afdac0f9fad0bcf74a34876c084d69e0a912f79f0228d1dc9667d41bbfdb7b
|
|
BLAKE2b-256 checksum How to use checksums |
016a9885aeb2ee4f4484aeaa17b298d65d57bde88d37f037b6191857ab5865db
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.2 CPython/3.11.2
|
Security for engression 1.0.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.