catacomb-ai / catacomb

The simplest machine learning library for launching UIs, running evaluations, and comparing model performance.

The simplest machine learning library for deploying prototypes, conducting quality assurance, and tracking production model performance.

Preview

Usage

Installation

Catacomb's Python library can be installed from the PyPi registry:

pip install catacomb-ai

To test installation, run catacomb:

Usage: catacomb [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  build
  push
  run

Required Files

The only file Catacomb expects is a system.py file that implements a class containing the __init__ and output() methods:

import catacomb

class UppercaseModel:
    def __init__(self):
        """Initializing system and loading dependencies"""
        self.variable = True

    def output(self, text):
        """Performing inference and returning a prediction"""
        return text.upper()
        
if __name__ == "__main__":
    catacomb.connect(UppercaseModel, 'TEXT')

Implementing the system interface allows Catacomb to auto-generate a UI for the system/model from the command line tool. Model hosting will fail unless all dependencies are defined within the current directory (i.e. a Pipfile or requirements.txt file is required).

Running Catacomb locally can be done by running python system.py.

Deployment

Uploading to the Catacomb hosting platform can be done by running:

catacomb upload

and following the command-line prompts to configure meta-data and example test cases.

External Dependencies

Additional external dependencies can be installed by specifying a catacomb.sh bash file to run on the created image. This file is detected during the catacomb upload build process.