https://github.com/elisim/piven

Official implementation of the paper "PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction" by Eli Simhayev, Gilad Katz and Lior Rokach.
https://github.com/elisim/piven

deep-learning prediction-intervals regression tensorflow uncertainty-estimation

Last synced: 2 months ago
JSON representation

Official implementation of the paper "PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction" by Eli Simhayev, Gilad Katz and Lior Rokach.

• Host: GitHub
• URL: https://github.com/elisim/piven
• Owner: elisim
• License: mit
• Created: 2020-10-08T10:11:37.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2023-03-12T05:51:21.000Z (almost 2 years ago)
• Last Synced: 2023-10-20T19:39:34.594Z (over 1 year ago)
• Topics: deep-learning, prediction-intervals, regression, tensorflow, uncertainty-estimation
• Language: Jupyter Notebook
• Homepage:
• Size: 5.26 MB
• Stars: 30
• Watchers: 2
• Forks: 4
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction

The official implementation of the paper ["PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction"](https://arxiv.org/abs/2006.05139)

by Eli Simhayev, Gilad Katz and Lior Rokach. 

## Update 21.6.22

Our paper has been accepted to Knowledge Based Systems 🙂 

https://doi.org/10.1016/j.knosys.2022.108685

 

![The PIVEN schematic architecture](piven_architecture.jpg)

## TL;DR

A neural network with a PIVEN output layer returns a point prediction as well as a lower and upper prediction interval (PI) for each target in a regression problem. The image below shows how the lower and upper PI change as we keep training the model:







We thank Jasper Ginn (@JasperHG90) from GoDataDriven ([godatadriven/piven](https://github.com/godatadriven/piven)) for the image 🙂

## Quickstart in Google Colab

A simple fast colab demo using Keras is included in [PIVEN_Demo.ipynb](https://colab.research.google.com/github/elisim/piven/blob/master/PIVEN_Demo.ipynb).

## Contents

```

├── age

│   ├── Bone age ground truth.xlsx --- RSNA Bong Age Ground-Truth

│   ├── get_age_data.sh --- Download dataset from kaggle

│   ├── main.py --- Run bone age experiment

├── imdb

│   ├── densenet.py

│   ├── generators.py

│   ├── get_imdb_data.sh --- Download dataset

│   ├── imdb_create_db.py --- Run after downloading the dataset

│   ├── main.py --- Run imdb age estimation experiment

│   ├── model.py

│   ├── subpixel.py

│   ├── tensorflow_backend.py

│   ├── train_callbacks.py

│   └── utils.py

└── uci

    ├── code

    │   ├── DataGen.py

    │   ├── DeepNetPI.py

    │   ├── alpha_experiment.py --- Run alpha experiment on UCI 

    │   ├── main.py --- Run UCI experiments

    │   ├── params_deep_ens.json --- deep ensembles hyperparameters

    │   ├── params.json --- piven and qd hyperparameters

    │   └── utils.py

    ├── get_song_dataset.sh --- Download Year Prediction MSD dataset

    └── UCI_Datasets

```

## Requirements

* pandas==0.25.2

* numpy==1.18.1

* matplotlib==3.0.3

* tensorflow==1.15.0

* keras==2.3.1

* xlrd==1.2.0

* scikit-learn==0.22

* tqdm==4.45.0

* opencv-python==4.2.0.34

To install requirements:

```setup

pip install -r requirements.txt

```

All experiments tested on Ubuntu 18.04 with Python 3.6.

## Acknowledgements

Our UCI experiments were inspired by Tim Pearce's implementation of [High-Quality Prediction Intervals for Deep Learning:

A Distribution-Free, Ensembled Approach](https://github.com/TeaPearce/Deep_Learning_Prediction_Intervals). Moreover, in 

IMDB age estimation experiment we used the preprocessing implemented in 

[SSR-Net: A Compact Soft Stagewise Regression Network for Age Estimation](https://github.com/shamangary/SSR-Net).

## Citing PIVEN

If you use PIVEN in your research please use the following BibTeX entry:

```BibTeX

@article{simhayev2022integrated,

  title={Integrated prediction intervals and specific value predictions for regression problems using neural networks},

  author={Simhayev, Eli and Katz, Gilad and Rokach, Lior},

  journal={Knowledge-Based Systems},

  volume={247},

  pages={108685},

  year={2022},

  publisher={Elsevier}

}

```