Skip to content

nuprl/TypeWeaver

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TypeWeaver

This is the code for the ECOOP 2023 paper: Do Machine Learning Models Produce TypeScript Types that Type Check?

The main branch of this repository is the latest development branch.

The ecoop2023-artifact branch is a snapshot of the repository that was used to prepare the ECOOP 2023 artifact.

  • DOI

Dependencies

  • Hardware: a GPU with at least 14 GB of VRAM
    • It is possible to skip the experiments that require a GPU
  • Software:
    • Linux
    • Python +3.6 and the tqdm package (pip install tqdm)
    • Optional, for dataset construction:
      • Node +16
      • git
    • Podman with the NVIDIA container toolkit
      • All other dependencies are managed with OCI images
      • Docker may be used instead of Podman, but has not been tested
      • It is possible to run without containers, but not recommended, as setting up the dependencies is very difficult

Makefile Targets

The Makefile has the following targets:

  • build: build the project

  • all: run everything

  • predict-all: run type prediction for all models

  • weave-all: run type weaving for all models

  • typecheck-all: run type checking for all models

  • predict, weave, typecheck: run type prediction, type weaving, or type checking, for the model specified by the MODEL variable

  • csv: generate summary CSVs

Makefile Variables

No containers

By default, containers are built and used to invoke the models. If dependencies are satisfied, it is possible to run without containers:

make NOCONTAINERS=true build
make NOCONTAINERS=true all

No GPU

Setting this variable will skip the InCoder experiments, which require a GPU:

make NOGPU=true all

Model

When running make predict, make weave, or make typecheck, the MODEL variable must be provided, e.g.:

make MODEL=DeepTyper predict

This will run type prediction for DeepTyper and no other model.

MODEL must be one of: DeepTyper, LambdaNet, or InCoder.

Number of processors

By default, make all will use all available processors on the machine. This can be configured:

make NPROC=8 all

The other targets only use a single processor.

Docker

By default, we use Podman. To use Docker (or another container implementation), set the DOCKER variable:

make DOCKER=docker build
make DOCKER=docker shell
DOCKER=docker ./run.sh

About

Artifact for the ECOOP 2023 paper: Do Machine Learning Models Produce TypeScript Types that Type Check?

Resources

Stars

Watchers

Forks

Packages

No packages published