model-card-printer 0.1.1

A streamlined approach to model card generation for machine learning models

Model Card Printer

The Model Card Printer simplifies the generation of model cards for machine learning models. Model Card Printer offers a standardized and structured approach to documenting machine learning models, while still providing flexibility for users to customize the document without being overly restrictive.

• Works with popular data visualization libraries such as Pandas and Plotly.
• Supports all text-based documentation in Markdown format.
• Tailor your model card to fit your needs and preferences.
• Supports exporting model cards in JSON, Pickle, and HTML formats.

Installation

Model Card Printer is available as a PyPI package. Install it using pip package manager:

pip install model-card-printer

Quickstart

1. Create documentation and data visualizations

Note: All visualizations generated here are for demonstration purposes. You could use Pandas to create your visualizations as well.

import plotly.express as px

# Create documentation (markdown format)
documentation_standalone = "# Model Card Demo\nHere is a paragraph related to the documentation.\n## Considerations\nThis demonstration is for anyone that wants to quickly start generating model cards with their trained ML models."

# Create data visualizations and related documentation
df = px.data.tips()
fig1 = px.box(df, y="total_bill")
fig2 = px.histogram(df, x="total_bill", color="sex", marginal="rug",
                         hover_data=df.columns)
fig3 = px.density_heatmap(df, x="total_bill", y="tip")

documentation_for_visuals = "# Quantitative Analysis\nThe visualizations presented are for demonstration purposes only."

# Create standalone data visualizations
df = px.data.iris()
fig_standalone = px.scatter_matrix(df,
    dimensions=["sepal_length", "sepal_width", "petal_length", "petal_width"],
    color="species", symbol="species",
    title="Scatter matrix of iris data set",
    labels={col:col.replace('_', ' ') for col in df.columns})
fig_standalone.update_traces(diagonal_visible=False)

2. Create model card

from model_card_printer import ModelCardPrinter

# Generate an empty model card
model_card = ModelCardPrinter.generate_card()

# Store all the information
model_card.add_custom_documentation(documentation_standalone, document_name = 'demo')
model_card.add_custom_visualization_collection({"demo fig 1": fig1, "demo fig 2": fig2, "demo fig 3": fig3},
                                               documentation = documentation_for_visuals,
                                               collection_name = "Demo visuals and text")
model_card.add_individual_custom_visualization(fig_standalone, "demo standalone fig")

# Save model card as HTML (default dark mode, else add parameter "is_dark_mode = False")
model_card.write_html("demo_model_card.html")

Model Card Export Formats

1. Output as JSON

# Saving to JSON file
model_card.write_json("model_card.json")

# OPTIONAL: Intermediate output
model_card_json = model_card.to_json()

2. Output as HTML

# Saving to HTML file
model_card.write_html("model_card.html")

# OPTIONAL: Intermediate output
model_card_html = model_card.to_html()

3. Output as Pickle object (not recommended unless there is a need to store the datasets inside the model card)

import pickle

with open('model_card.pkl', 'wb') as file:
    pickle.dump(model_card, file)

Model Card Import Formats

1. Load from JSON file

model_card = ModelCardPrinter(file_path = "model_card.json").load_from_json()

2. Load from Python dictionary

dict_model_card = previous_model_card.to_dict()
model_card = ModelCardPrinter(data_dict = dict_model_card).load_from_dict()

Tutorials

For more in-depth demonstrations on model-card-printer package, refer to the provided examples.

• Creating a model card using the Breast Cancer dataset
• Create a custom model card