Le Board (beta)

Le Board, a simple leaderboard generator for ranking your algorithms experiments.

Install

Install with pip and configure with init, linking a Google account is the recommended choice.

pip install leboard

leboard init

sketch of leboard init

Choose an option for Data Handler (1):

1. Google Firebase (recommended)
2. Local file system (no shared options)

if option 1)

• Choose a name for Firebase project (leboard)
• Create a Firebase project at https://console.firebase.google.com/
• Get the gcloud credential at https://console.cloud.google.com/iam-admin/serviceaccounts/project
• Access the page Database -> activate the Firestore database

if option 2)

• Choose a path to store your board (~/.leboard):

Usage

import leboard

letask = leboard.TaskBoard("MNIST")

# other members will have access to the MNIST task too
# (restrict a members definition only when a task is set)
letask.members(["rob@gmail.com", "jane@mail.com"])

# do your stuff
# ...

experiment = letask.Entry()

experiment.set("accuracy", accuracy)
experiment.set("loss", loss)
experiment.set("parameters", params, hide=True)
experiment.set("confusion_image", confusion_image)

# commit will save to DataHandles available/configured
# (proposed is local simple file storage, and shared google firebase)
experiment.commit()

Leaderboard

Show all experiments with links to detailed resources, this can open a page hosted in firebase.

• keep experiment data in order of creation

letask.leaderboard("MNIST", "accuracy")

Advanced?

Resource Typing?

Entry can have different types, this can be usefull for handling resources like image, or HTML pages, to show and interact with them.

from leboard import types as lbt

experiment.set("accuracy", accuracy)
experiment.set("loss", loss)
experiment.set("confusion_image", lbt.Image(confusion_image))
experiment.set("notebook_html", lbt.HTML(notebook_html))

Committing an Entry could automatically save a copy of the current notebook state to the DataHandler.