Working directory for Learning One-of-Many Solutions for Combinatorial Problems in Structured Output Spaces. Use this repo for replicating experiments in paper exactly. For cleaner version of code and documentation please refer https://github.com/dair-iitd/1oML.
We experiment with 2 base models: Neural Logic Machines (NLM) and Recurrent Relational Networks (RRN).
Our training code has been adapted from google/neural-logic-machines and uses Jacinle python toolbox.
Before installation you need
- python 3
- numpy
- tqdm
- yaml
- pandas
- PyTorch >= 1.5.0 (should probably work with >=1.0 versions, tested extensively with >=1.5)
Clone this repository:
git clone https://github.com/dair-iitd/1oML --recursive
Install Jacinle included in third_party/Jacinle. You need to add the bin path to your global PATH environment variable:
export PATH=<path_to_1oML>/third_party/Jacinle/bin:$PATH
Create a conda environment for 1oML, and install the requirements. This includes the required python packages
from both Jacinle and NLM. Most of the required packages have been included in the built-in anaconda package:
conda create -n nlm anaconda
conda install pytorch torchvision -c pytorch
Install dgl for RRN.
- Download datasets from this drive link.
- Run experiments from
scripts/shell_commands/[nqueens|futo|sudoku]_e2e_[baselines|ccloss|minloss|Iexplr|selectR].sh