This repo contains python implentations of Efficient Nonmyopic Active Search, ICML 2017 and Efficient Nonmyopic Batch Active Search, NeurIPS 2018.

Active search is an active learning paradigm which seeks to sequentially inspect data so as to discover members of a rare, desired class. The labels are not known a priori but can be revealed by querying a costly labeling oracle. The goal is to design an policy to sequentially query(observe) points to find as many valuable points as possible under a labeling budget. The Bayesian optimal policy for Active Search was derived in Efficient Nonmyopic Active Search, ICML 2017, which automatically and dynamically balances exploration and exploitation consistent with the remaining budget, without relying on a parameter to control this tradeoff.

Below is an example of Efficient Nonmyopic Active Search(ENS) running on a toy 2-d problem with a budget of 300 queries:

For a Matlab implementation of ENS written by Roman Garnett and Shali Jiang, look here. A 3-minute video introducing Efficient Nonmyopic Batch Active search can be found here

This code can run on Unix,MacOS, Windows. Simply clone the repository, install requirements, and run the demo like this:

$ git clone https://github.com/MatthewGleeson/ActiveSearch.git
$ pip install -r requirements.txt
$ python3 scriptToRun.py

Change parameter settings in the script file to try different datasets and policies!

The code has been tested on MacOS High Sierra 10.13.6 with Python 3.6.0.