Introduction

This page describes the experiment system with Picsell.ia.

Here you will learn the purpose of experiments, what it is made of and what it looks like.

The experiment system

When you have created a project, added your team members, now is the time to start logging everything you do, store every asset you create so you and your team can always visualize, access and use it anywhere, anytime.

The experiments are atomic objects belonging to your projects that are composed of the following properties/assets :

  • Name

  • Description

  • Status

  • Data Assets

  • File Assets

  • Logs (Telemetry)

Name (60 characters max.)

The name of your experiment must be unique within a single project, it is displayed when listing all your experiments and allows you to retrieve your experiment with our Python SDK.

Description (1000 characters max.)

This text is used to shortly describe your experiment. How is it different from the others ? What parameter has changed ? It is displayed when listing your experiments so you can find what you are looking for before diving in each experiment details.

Status (60 characters max.)

This parameter is used to describe the current stat of your experiment. It is displayed when listing your experiments and is also used by the Python SDK for example to retrieve an experiment that needs to be run.

You can set any value for this property, but please keep in mind that some name are reserved and already used by Picsell.ia to automate stuff, so be aware when using a value within the following list :

  • created (default status when creating an experiment in our UI)

  • launched (default status when launching a managed training with Picsell.ia)

  • failed (default status when a managed training returned an error exit code)

  • succeeded (default status when a managed training returned a success exit code)

Assets

File Assets

You can store any file needed for and from your experiments on Picsell.ia such as :

  • Checkpoint files

  • Weights

  • Inference results

  • Config files

  • ...

All those files can be grouped in two main categories : small files (<5Mb) and large files (>5Mb) A file asset is composed of the following properties that you can edit :

  • name (100 characters max.), helpful to retrieve or visualize your asset with the SDK or on the platform

  • object_name (200 characters max.), is the path to the file on our object storage

  • large (Boolean) is set to True if the file-size is > 5Mb

  • last_update, the date with time of the last update of the file

  • date_created, the creation date of the file

Data assets

To be able to track your experiment, we enable you with the ability to send data in a dictionnary format so it can be displayed in the following plot :

  • Array

  • Line/Area Chart

  • Bar Chart

  • and many more coming soon

Alike file assets, a data asset is composed of the following properties :

  • name (100 characters max.), helpful to retrieve or visualize your asset with the SDK or on the platform

  • data, is a dictionnary where your data will be stored, it has to be organized in a particular way to be displayed properly on our platform (check doc here)

  • last_update, the date with time of the last update of the file

  • date_created, the creation date of the file

Logs

Logs are a special assets of your experiments, they are intended to store the stdout of your algorithms, you can then visualize the output of any program running on a remote instance in real time in this page of the platform.