AutoML

Attribute	Value
pretty_name	Automated Deep Learning
annotations_creators
language_creators
languages
licenses	unknown
multilinguality
size_categories	10k<n<100k
source_datasets
task_categories
task_ids
paperswithcode_id

Dataset Description

• Paper: Analysis of the AutoML Challenge Series 2015–2018

• Relevant Links:

  • Data

  • Challenge webpage

• Licenses: Unknown

Dataset Summary

Tabular data containing float elements.

Download and prepare data

The dataset can be loaded directly via the squirrel Catalog API. Make sure that squirrel-dataset-core is installed via pip, which will register this dataset. Use the following code to load the data:

from squirrel.catalog import Catalog
plugin_catalog = Catalog.from_plugins()
it = plugin_catalog["helena"].get_driver().get_iter(split="train")

from squirrel.catalog import Catalog
plugin_catalog = Catalog.from_plugins()
it = plugin_catalog["jannis"].get_driver().get_iter(split="train")

Dataset Structure

Data Instances

A sample from the Helena training set is provided below:

{
    'features': [
        0.200384,
        0.660417,
        0.4375,
        0.38136,
        0.531051,
        0.543844,
        0.378399,
        0.025277,
        0.306467,
        123.23,
        105.248,
        95.8235,
        51.416,
        50.8848,
        47.8058,
        0.735954,
        0.876967,
        1.2111,
        69.2957,
        3.7954,
        5.33528,
        12.7654,
        2.49029,
        4.30002,
        0.590964,
        -0.0334237,
        0.394317
    ],
    'class': 9
}

Dataset Schema

All features are continuous floats. There are in total 100 classes to predict. Test and validation data do not contain class labels

Data Splits

name	samples
Helena train	65,196
Helena test	18,628
Helena valid	9,314
Jannis train	83,733
Jannis test	9,851
Jannis valid	4,926