NannyML Datasets

This Python package contains our curated datasets, used for testing and product demos.

We provide datasets for the following problem types: binary classification, multiclass classification and regression.

Anatomy of a Dataset

Each Dataset has reference and monitoring properties. Each of these exposes the following properties:

• data: access the full dataset as a pyarrow.Table
• predictions: access the model predictions as a numpy.ndarray
• predicted_probabilities: access the model's predicted probablilities. Only available for classification datasets. For binary classification datasets this will be a single
• targets: access the model targets as a numpy.ndarray
• timestamps: access the model timestamps as a numpy.ndarray
• categorical_features: access the model's categorical features. Loop over tuples containing the column name and its values.
• continuous_features: access the model's continuous features. Loop over tuples containing the column name and its values.
• features: access the model's features. Loop over tuples containing the column name and its values.

If any of these properties are not available, trying to access them will raise an AssertionError.

Example usage

from nannyml_dataset.binary_classification import synthetic_car_loan  # Import the dataset

print(synthetic_car_loan.reference.timestamps)  # Access some reference property
print(synthetic_car_loan.monitoring.timestamps)  # Access some monitoring property

for name, values in synthetic_car_loan.reference.categorical_features:  # Loop over reference categorical features
    print(f"{name}\t\t{values}")  # You can do more useful stuff here, like setting up a univariate covariate shift monitor!

Available datasets

Binary Classification

Dataset	Synthetic	Description
synthetic_car_loan	yes	A synthetic dataset describing a model that predicts defaulting a loan for a car.
hotel_booking	no	A dataset describing a model that predicts booking cancellation in a hotel context.

Multiclass classification

Dataset	Synthetic	Description
synthetic_credit_card	yes	A synthetic dataset describing a model that predicts a class of credit card (upmarket, highstreet, prepaid).
satellite_imagery	no	A dataset describing a model that classifies an image tile as earth, water, desert, ...

Regression

Dataset	Synthetic	Description
synthetic_car_price	yes	A synthetic dataset describing a model that predicts the price of a second-hand car.