ML platform


Keywords
data, science, machine, learning, pipeline, workflow, experiments, collaboration, data-science, distributed, machine-learning, plynx, reproducibility
License
Apache-2.0
Install
pip install plynx==1.7.9

Documentation

Plynx

Plynx

Website shields.io Website shields.io Website shields.io

Website and demo: plynx.com.

Docs: docs.

PLynx is a domain agnostic platform for managing reproducible experiments and data-oriented workflows.

Features

Workflow Editor

Interactive User Interface. You can clone successful experiment and reuse it or create one from scratch. PLynx manages history of the experiments and they can be reproduced.

interactive graph editor

Operations editor

Operations can be customized independently from the platform. Users can define their own Operations or reuse existing ones.

online editor

Monitor progress

Track the progress of the experiment. Each of intermediate operations produce results that you can inspect.

monitor progress

Preview the results

View the results right in the browser.

results preview

Scalable architecture

Execution Engine is based on scalable pub/sub model. Each Worker performs their jobs independently from each other and can publish subtasks back to the queue. Executers are plugins themselves and can support multiple scenarios from "compile to binary code" to "deploy and serve" to "run in a cluster using as many distributed workers as possible".

Scalable architecture

Requirements

Plugins work on python3. User Interface is based on React. PLynx is using MongoDB as a primary metadata storage. In order to meet diverse data storage requirements, its own storage plugins to store the artifacts. It supports multiple data storages such as AWS S3, Google Cloud Storage and traditional filesystems.

In order to reduce complexity we recommend to install docker and run make command to start local cluster.

Get started

Usage

Make sure you install docker first. Get started with Docker

tl;dr

git clone https://github.com/plynx-team/plynx.git   # Clone the repo

cd plynx

cp template_config.yaml config.yaml                 # Make a copy of a config
make up                                             # to start production services

Then go to http://localhost:3001

By default it will start the following services:

  • MongoDB instance
  • PLynx User Interface
  • Backend
  • Several workers

Other make commands:

  • make build - build all docker images.
  • make run_tests - build docker images and run the tests.
  • make up - run the services locally.
  • make dev - run developer version of PLynx.

Config

Most of the parameters can be set in command line when the services are called. For example:

$ plynx exec --help
usage: -c exec [-h] [-v] -f FILENAME [--storage-prefix STORAGE_PREFIX]

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         Set logging output more verbose
  -f FILENAME, --filename FILENAME
                        Path to file
  --storage-prefix STORAGE_PREFIX
                        Storage prefix

But we recommend to store the config in a separate file.

Plynx config location is can be set in env variable PLYNX_CONFIG_PATH. Default value is ./config.yaml.

Install PLynx with dev requirements

Run:

pip install -e .[all]

pre-commit install

External links