PyBoard

Little app for tracking ML models learning progress. Built using React, Apollo (GraphQL) and Node.

Getting started:

$ git clone https://github.com/c0z0/pyboard.git
$ cd pyboard
$ yarn install
$ yarn dev
  or
$ npm install
$ npm run dev

Using the Python client example

import pyboard # Copy pyboard.py in python project root
  
url = 'http:localhost:3000'
  
pyboard_client = pyboard.BoardClient(url)
  
mnist_experiment = pyboard.create_experiment('MNIST')
  
loss_logging = mnsit_experiment.createScalar('Loss')
accuracy_logging = mnsit_experiment.create_scalar('Accuracy')
  
#                         (...)
  
# In training loop:
loss_logging.add_value(loss, batchN)
accuracy_logging.add_value(acc, batchN)

Python client API

pyboard.BoardClient(url)

Creates a pyboard client.

Params:

url Url of the pyboard server

client.create_experiment(name)

Creates an experiment and returns it.

client.get_experiment(id)

Finds experiment by id and returns it.

 experiment.create_scalar(name)

Creates a scalar and returns it

experiment.get_scalar(name, [id])

Finds scalar by name or id and returns it.

experiment.name # Name of the experiment
experiment.id # Id of the experiment

scalar.add_value(value, time, batch_wait_time=10)

Adds a value to the scalar

Params:

value Value to add time Value to show on the time axis (Eg: epoch_number) batch_wait_time Time (in seconds) between request batches.

Todo:

• Delete / Clear experiments and scalars
• Add a persistance layer (probably both fs and db) (currently the data is saved in memmory)