Skip to content

This is the official repository for "Explanatory Learning: Beyond Empiricism in Neural Networks".

Notifications You must be signed in to change notification settings

gladia-research-group/explanatory-learning

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Explanatory Learning

This is the official repository for "Explanatory Learning: Beyond Empiricism in Neural Networks" [arXiv][PDF].

Datasets

Download the datasets with the following links:

After downloading, move the datasets inside the directory data, located in the repository root folder.

Usage

First, clone the github repository

git clone https://github.com/gladia-research-group/explanatory-learning.git

and move the current directory int the repository root:

cd explanatory-learning

Dependencies

This repository has the following dependencies:

  • Python 3.8
  • PyTorch 1.7 (look here)

All other dependencies can be easily installed by running:

pip install -r requirements.txt

Training

It is possible to train the models described in the paper with the following commands. First, while inside the root directory, you can start the training procedure for the models described in the paper with the following command:

python explanatory_learning/learning/training.py --config-file "config.json"

This will create a folder in the repository root directory named training_results. This folder is necessary in order to run the experiments in the paper.

NOTE: It is possible to change the configuration file config.json (located in the repository root folder) to modify the training procedure.

Experiments

In order to execute the experiments reported in the paper you can use the following commands.

  • Rule Induction Task: For the results presented in Table 1, run:
     python explanatory_learning/experiments/rule_induction.py --config-file "configs/config.json"
  • World Labeling Task: For the results presented in Table 2, run:
    python explanatory_learning/experiments/world_labeling.py --config-file "configs/config.json"
  • Rank Analysis: For the results presented in Table 3, run:
    python explanatory_learning/experiments/rank_analysis.py --config-file "configs/config.json"
  • Interpreter Evaluation: For the results presented in Figure 4, run:
     python explanatory_learning/experiments/interpreter_analysis.py --config-file "configs/config.json"

NOTE: Each command will generate a respective folder in the repository root directory containing the results in json format. In particular, the file

python explanatory_learning/experiments/<X>.py --config-file "configs/config.json"

will generate the directory <X>_results.

About

This is the official repository for "Explanatory Learning: Beyond Empiricism in Neural Networks".

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages