Implementation:Online ml River Datasets Insects

Knowledge Sources	Online_ml_River
Domains	Online_Learning, Datasets, Multi_Class_Classification, Concept_Drift
Last Updated	2026-02-08 16:00 GMT

Overview

Concrete dataset for multi-class classification with concept drift provided by the River library.

Description

The Insects dataset has different variants specifically designed for concept drift evaluation. The number of samples and the difficulty change from one variant to another. The number of classes is always 6, except for the "out-of-control" variant which has 24 classes.

Available variants:

• abrupt_balanced
• abrupt_imbalanced
• gradual_balanced
• gradual_imbalanced
• incremental_abrupt_balanced
• incremental_reoccurring_balanced
• incremental_balanced

The default variant is "abrupt_balanced" with 52,848 samples and 33 features.

Usage

This dataset is useful for:

• Evaluating concept drift detection algorithms
• Testing stream learning algorithms on non-stationary data
• Comparing performance on balanced vs imbalanced scenarios
• Benchmarking classification algorithms with different drift patterns

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/datasets/insects.py

Signature

class Insects(base.RemoteDataset):
    def __init__(self, variant="abrupt_balanced"):
        if variant not in self.variant_configs:
            variants = "\n".join(f"- {v}" for v in self.variant_configs)
            raise ValueError(f"Unknown variant, possible choices are:\n{variants}")

        config = self.variant_configs[variant]
        n_samples = config["n_samples"]
        size = config["size"]
        url = config["url"]
        filename = config["filename"]
        n_classes = 24 if variant == "out-of-control" else 6

        super().__init__(
            n_classes=n_classes,
            n_samples=n_samples,
            n_features=33,
            task=base.MULTI_CLF,
            url=url,
            size=size,
            unpack=False,
            filename=filename,
        )
        self.variant = variant

    def _iter(self):
        cols = [f"f{i}" for i in range(1, 34)] + ["class"]
        return stream.iter_csv(self.path, target="class", fieldnames=cols)

Import

from river import datasets
dataset = datasets.Insects()  # default: abrupt_balanced
# Or specify a variant:
dataset = datasets.Insects(variant="gradual_balanced")

I/O Contract

Inputs

Name	Type	Required	Description
variant	str	No	Variant name (default: "abrupt_balanced")

Outputs

Name	Type	Description
iter()	tuple(dict, str)	Yields (features_dict, target) pairs where features are f1-f33

Dataset Properties

Property	Value (default variant)
Number of samples	52,848 (varies by variant)
Number of features	33
Number of classes	6 (24 for "out-of-control")
Task	Multi-class classification
Format	CSV

Variants

Variant	Samples	Size (bytes)	Description
abrupt_balanced	52,848	14,151,769	Balanced classes with abrupt drift
abrupt_imbalanced	355,275	94,893,622	Imbalanced classes with abrupt drift
gradual_balanced	24,150	6,474,831	Balanced classes with gradual drift
gradual_imbalanced	143,323	38,339,554	Imbalanced classes with gradual drift
incremental_abrupt_balanced	79,986	21,421,452	Balanced with incremental abrupt drift
incremental_reoccurring_balanced	79,986	21,433,047	Balanced with reoccurring drift
incremental_balanced	57,018	15,258,997	Balanced with incremental drift

Usage Examples

from river import datasets

# Use default variant
dataset = datasets.Insects()
for x, y in dataset:
    print(x, y)
    break

# Use specific variant for gradual drift
dataset = datasets.Insects(variant="gradual_balanced")
for x, y in dataset:
    print(x, y)
    break

References

• USP DS repository
• Souza, V., Reis, D.M.D., Maletzke, A.G. and Batista, G.E., 2020. Challenges in Benchmarking Stream Learning Algorithms with Real-world Data. arXiv preprint arXiv:2005.00113. [1]

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment