Differentiable ILP for Structured Examples

An implementation of the paper: Differentiable Inductive Logic Programming for Structured Examples

Requirement

• python >= 3.6.3

See requirements.txt for more details.

Installation

pip install -r requirements.txt

How to use

DEMO

A simple demonstration is available in demo.ipynb.

Or on the terminal,

python -m experiments [type] [name] [lr] [epoch] [m:target program size] [T:infer_step] [--nosie_rate]

• [type]: normal, noise, step, softor
• [name]: member, plus, append, delete, subtree

For example,

python -m experiments normal member 1e-2 3000 2 4 --noise_rate 0.05
python -m experiments normal plus 1e-2 3000 3 8 --noise_rate 0.05
python -m experiments normal append 1e-2 3000 3 4 --noise_rate 0.05
python -m experiments normal delete 1e-2 3000 2 4 --noise_rate 0.05
python -m experiments normal subtree 1e-2 3000 3 4 --noise_rate 0.05

Experiments

For the noise experiments, for example,

python -m experiments noise member 1e-2 3000 2 4

The results are summarized in imgs folder.

For the clause-generation experiments, for example,

python -m experiments step delete 1e-2 3000 2 4

The results are summarized in imgs folder.

For the soft program-composition experiments, for example,

python -m experiments softor append 1e-2 3000 3 4

The results are summarized in results folder.