Picsellia
  • Picsellia
  • Getting started
    • Start using Picsellia
    • Create, annotate, review a Dataset
    • Create a new Dataset Version with merged labels
    • Train a custom Object Detection model
    • Train a custom Classification model
    • Deploy model in production (Tensorflow only)
    • Feedback loop - Send predictions from models to Datalake or Datasets
  • Data Management
    • Upload assets to your Lake
    • Edit Tags for Pictures
    • Create a Dataset
    • Add data to a Dataset
    • Create a new Dataset version
    • Configure your Labels
    • Import annotation from other Dataset version
  • Experiment Tracking
    • Initialize an experiment
    • Checkout an experiment
    • Log your results to Picsell.ia
    • Store your files to Picsell.ia
    • Evaluate your models
    • Retrieve and update your assets
    • Publish your model
    • Namespace
  • Hyperparameter tuning
    • Overview
    • Install Picsell CLI
    • Config
    • Launch your Hyperparameters tuning
  • Models
    • Model HUB
  • References
    • API Reference
    • Python SDK Reference
    • Python Training Reference
  • Organization
  • Website
Powered by GitBook
On this page

Was this helpful?

  1. Experiment Tracking

Publish your model

Here we will see how to publish a version of your experiment so it can be used outside of the project scope and deployed for production use

PreviousRetrieve and update your assetsNextNamespace

Last updated 4 years ago

Was this helpful?

Once you are happy with one of your experiment after analyzing the evaluation metrics or sample predictions, you can publish it into a new frozen entity that we call a model.

Once your experiment is frozen into a model, you can then deploy it, use it live in the playground, or use it as a source for other experiments that can be in other projects (remember the method ?)

To freeze your experiment into a model, you can use the following method :

from picsellia.client import Client

api_token = '4a54b5d45e45f4c454b54dee5b54bac4dd4'
project_token = '9a7d45b4c-691d-4c3a-9972-6a22b1dcd6f'

experiment = Client.Experiment(
    api_token=api_token,
    project_token=project_token
    )
experiment.checkout(
    name='my_finished_experiment'
    )
experiment.publish('my_awesome_model')

As we can see, the only thing we have to specify is the name of the model instance we want to create. When publishing, you will automatically clone the file assets of the experiment such as :

  • checkpoints

  • trained model

  • .config

  • ...

checkout