SKOPS

Last updated Jan 3, 2023

SKOPS is python library that allow to share your model huggingface.

• 3 files
  • README.md
  • config.json
  • model file

To use model the huggingface hub needs some info, which is stored in metadata in README.md or config.json

https://huggingface.co/docs/hub/model-cards

{
    "sklearn": {
	"columns": [
	    "petal length (cm)",
	    "petal width (cm)",
	    "sepal length (cm)",
	    "sepal width (cm)",
	],
	"environment": ['scikit-learn="1.1.1"', "numpy"],
	"example_input": {
	    "petal length (cm)": [1.4, 1.4, 1.3],
	    "petal width (cm)": [0.2, 0.2, 0.2],
	    "sepal length (cm)": [5.1, 4.9, 4.7],
	    "sepal width (cm)": [3.5, 3.0, 3.2],
	},
	"model": {"file": "model.pkl"},
	"task": "tabular-classification",
    }
}

The config file is created automatically by “skops.hub_utils.init()” You almost never need to create or touch this file manually, and it’s created when you call skops.hub_utils.init(). ``

# contents of train.py
...
hub_utils.init(model, dst=local_repo)
hub_utils.add_files(__file__, dst=local_repo)  # adds train.py to repo
hub_utils.push(...)

https://huggingface.co/docs/hub/model-cards#model-card-metadatahttps://github.com/huggingface/hub-docs/blob/main/modelcard.md?plain=1https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md

Card.add_plot() method. For tables, you can pass either dictionaries with the key being the header and the values being list of row entries, or a pandas DataFrame; use the Card.add_table() method for this.

The template also contains the following sections that are automatically generated by skops.

“Hyperparameters”: Hyperparameters of the model.

“Model Plot”: A diagram of the model, most relevant in case the model is a complex scikit-learn Pipeline.

https://github.com/skops-dev/skops/issues/248https://skops.readthedocs.io/en/stable/model_card.htmlhttps://github.com/skops-dev/skops/pull/257

https://github.com/skops-dev/skops/pull/203/commits/68e474822906490e0013edc1d75014df10dd7e3c

To create a custom template user can define a template using dict.

User can add metrics plots using convienence method.

How can a user add certain section from default template.

currently, The I need section names but the problem i self.add(**SKOPS_TEMPLATE) , or specify certain section from the

then I can use self.add_model_plot() self.add_hyperparams() self.add_get_started_code()

to add other section to the model.

In this way I can use custom template, also use sections from the default templates.