Vampire+ML

Machine learning extensions for Vampire

Experiments in Neptune.ai: filipbartek/vampire-ml

Setup

To install TensorFlow for GPU with compatible versions of its dependency cudatoolkit:

conda install tensorflow=*=gpu_*

Quick start

Before running the experiments for the first time, it is necessary to build the Vampire prover located in the Git submodule vampire.

Before running an experiment, configure the following environment variables:

• DGLBACKEND=tensorflow
• XDG_CACHE_HOME: Path to a cache directory. System default: $HOME/.cache
• VAMPIRE: Path to a Vampire prover binary. Example: vampire/build/release/bin/vampire
• TPTP: Path to a TPTP library directory. Example: $HOME/TPTP-v7.4.0

Call the module qustions to run an experiment. Example calls:

# Train a predicate precedence recommender using a dataset of 1000000 examples
python -m questions questions.max_count=1000000 questions.randomize=[predicate] symbol_type=predicate

# Evaluate a predicate precedence recommender stored in the checkpoint "outputs/2021-02-16/12-28-14/tf_ckpts/epoch/weights.00289.tf" on all validation problems
python -m questions questions.max_count=1000000 questions.randomize=[predicate] symbol_type=predicate restore_checkpoint=outputs/2021-02-16/12-28-14/tf_ckpts/epoch/weights.00289.tf epochs=0 solver_eval.start=-1 solver_eval.iterations=5 solver_eval.problems.train=0 solver_eval.problems.val=null

# Print the supported parameters
python -m questions --help