Auto-WEKA : Sample Datasets

Below are some sample datasets that have been used with Auto-WEKA. Each zip has two files, test.arff and train.arff in WEKA's native format. To use these zip files with Auto-WEKA, you need to pass them to an InstanceGenerator that will split them up into different subsets to allow for processes like cross-validation.

To perform 10 fold cross-validation with a specific seed, you can use the following line for your instanceGeneratorArgs that you pass to the ExperimentConstructor:

numFolds=10:seed=1

When running an Auto-WEKA wrapper, you can then use the following 10 lines as an instanceString:

seed=1:numFolds=10:fold=0
seed=1:numFolds=10:fold=1
seed=1:numFolds=10:fold=2
seed=1:numFolds=10:fold=3
seed=1:numFolds=10:fold=4
seed=1:numFolds=10:fold=5
seed=1:numFolds=10:fold=6
seed=1:numFolds=10:fold=7
seed=1:numFolds=10:fold=8
seed=1:numFolds=10:fold=9

See the manual provided with Auto-WEKA for more details on how to chain InstanceGenerators together.

Data Sets

Name/Source	Auto-WEKA Zip	Size
Dexter	dexter.zip	311K
German Credit	creditg.zip	18K
Dorothea	dorothea.zip	3.3M
Yeast	yeast.zip	13K
Amazon	amazon.zip	2.3M
Secom	secom.zip	2.0M
Semeion	semeion.zip	51K
Car	car.zip	5.6K
Madelon	madelon.zip	1.6M
KR-vs-KP	krvskp.zip	17K
Abalone	abalone.zip	57K
Wine Quality	winequalitywhite.zip	74K
Waveform	waveform.zip	338K
Gisette	gisette.zip	12M
Convex	convex.zip	4.2M
CIFAR-10-Small	cifar10small.zip	67M
MNIST	mnist.zip	23M
Rot. MNIST + BI	mnistrotationbackimagenew.zip	114M
Shuttle	shuttle.zip	425K
KDD09-Appentency	kddcup09appetency.zip	9.2M
CIFAR-10	cifar10.zip	199M

Raw SMAC Data

We have been asked to provide the raw state files for the SMAC runs on these datasets (Note that they are not using the same version of Auto-WEKA as in the KDD paper). Raw Data (1.2GB)