Skip to content

Jupyter Models

Introduction

The BOSS platform provides JupyterLab as a tightly coupled part of a user’s workflow. Reaching JupyterLab can be done easily from an orange button at the top right of the GUI landing page (after logging in).

image

Once there, you can begin your work in notebooks!

Save a cell as a BOSS Model

After you’ve defined a model that you like (with the help of the python client - see python client documentation), it can be saved as a BOSS Model object just as if it had been uploaded through the GUI. This is accomplished by selecting your notebook cell, clicking the gear icons at the top right labelled “property inspector”, and adding Cell Metadata. Note that you do not need to update the Notebook Metadata section.

image

The required metadata is a JSON block which has a few components:

  • author_email: Email of the author
  • author_name: Name of the author
  • model_description: A description of the model
  • model_framework: One of simple, advanced, or federated.
  • model_name: A name for your BOSS Model.
  • model_type: One of tabular_classification, image_classification, text_classification, or regression.
  • project: A Project ID of the project you would like to put the model in. This can be retrieved from the GUI on the landing page under the triple dots.
  • tag: A tag unique to your cell. A notebook may contain multiple tagged models. The tags must have one of these prefixes, selected based on your model library:
    • TensorFlow: boss_model_tf_
    • PyTorch: boss_model_pt_
    • XGBoost: boss_model_xgb_
    • Sklearn: boss_model_skl_
    • Federated: boss_model_fl_

Here’s an example block creating an xgboost model:

{
    "author_email": "exampleuser@fake-email.com",
    "author_name": "Example",
    "model_description": "Testing notebook parsing.",
    "model_framework": "simple",
    "model_name": "JupyterFeb15th",
    "model_type": "tabular_classification",
    "project": "xIkNJn4BlBAQycdMgZzc",
    "tag": "boss_model_xgb_1"
}

After this metadata is added and saved, any changes made to the notebook cell will auto-update the BOSS Model object in the platform! The BOSS Model can be used for training just like a normal BOSS Model object. You will see it in the GUI denoted with a jupyter symbol. You may preview the code by clicking the eye icon, or jump directly to the related notebook back in jupyterlab by clicking the icon on the left (square with diagonal arrow)!:

image