Coactive Critiquing for Constructive Preference Elicitation

A python3 implementation of Coactive Critiquing for preference elicitation.

Coactive critiquing extends coactive learning with support for example critiquing interaction.

Please see our paper:

Stefano Teso, Paolo Dragone, Andrea Passerini. "Coactive Critiquing: Elicitation of Preferences and Features", accepted at AAAI'17, 2017.

Requirements

The following packages are required:

• numpy
• matplotlib
• pymzn, tested with version 0.10.7
• minizinc, tested with version 2.0.13
• gecode, tested with version 4.4.0
• cvxpy, tested with version 0.4.3

Usage

Type:

 $ ./main --help

to get the full list of options.

To run the synthetic experiment, type:

 $ ./main canvas ${method} -U 20 -T 100 -S ${sparsity} -E 0.1 -s 0 -d -W users/canvas_${sparsity}.pickle

where ${method} can be:

• pp-attr for pure Coactive Learning (fixed feature space) over the base features only
• pp-all for pure Coactive Learning (fixed feature space) over the full feature space
• cpp for Coactive Critiquing (dynamic feature space acquisition) and ${sparsity} is the degree of sparsity. The values used in the paper are 0.2 (sparse case) and 1.0 (non-sparse case).

Similarly, to run the travel planning experiment, type:

 $ ./main travel ${method} -U 20 -T 100 -S ${sparsity} -E 0.1 -s 0 -d -W users/travel_${sparsity}_tn_10.pickle

Funding

The project is supported by the CARITRO Foundation through grant 2014.0372.