Benchmarking Tests

Many of these tests were derived from the Machine Learning Repository, at the UC Irvine Center for Machine Learning and Intelligent Systems. This is a commonly used source of benchmarking data.

The below INQLE tests were run on a 1.6 GHz laptop computer running Windows XP.

Date	Test	Benchmark Results	INQLE Results	INQLE Details	Comment
11 Feb 2009: INQLE v 0.2.8, using the SiSSS Sampling Algorithm	Breast Cancer Wisconsin (Diagnostic) Data Set	"best predictive accuracy obtained using one separating plane in the 3-D space of Worst Area, Worst Smoothness and Mean Texture. Estimated accuracy 97.5% using repeated 10-fold crossvalidations. Classifier has correctly diagnosed 176 consecutive new patients as of November 1995."	INQLE v0.2.8 had 96.1% accuracy, in predicting the diagnosis (of cancer or not), using these attributes: Worst Radius, Worst Texture, and Mean Concave Points	Setup & import of data: 10 minutes Configure custom sampler to predict breast tumor diagnosis, which will randomly select 3 attributes to test their correlation with the diagnosis: 5 minutes Run the experimenter agent: Completed 5754 random experiments in 6 hours. (3.7 seconds per experiment)	This classification experiment tests INQLE's performance in selecting among 30 numeric attributes, which are highly correlated with the label (the attribute being predicted).
12 Feb 2009: INQLE v 0.2.8, using the SiSSS Sampling Algorithm and RapidMiner Nearest Neighbor regression learner	Computer Hardware Data Set	Published correlations: MCYT: -0.3071 MMIN: 0.7949 MMAX: 0.8630 CACH: 0.6626 CHMIN: 0.6089 CHMAX: 0.6052	INQLE correlations: MCYT: 0.6218 MMIN: 0.7763 MMAX: 0.8393 CACH: 0.7223 CHMIN: 0.7239 CHMAX: 0.2879	Setup & import of data: 10 minutes Configure custom sampler to relative CPU performance, which will randomly select 1 attribute to test its correlation with the label: 1 minutes Run the experimenter agent: Completed 37 random experiments in 51 seconds. (1.4 seconds per experiment)	This regression experiment tests INQLE's ability to employ regression learning algorithm, using a single numeric attribute, using a small data set.
18 Feb 2009: INQLE v 0.2.8, using the SiSSS Sampling Algorithm and RapidMiner Decision Tree learner	Iris Class Data Set	Published correlations: sepal length: 0.7826 sepal width: -0.4194 petal length: 0.9490 petal width:0.9565	INQLE correlations: sepal length: 0.7460 sepal width: 0.3996 petal length: 0.9654 petal width:0.9679	Setup & import of data: 10 minutes Configure custom sampler to predict iris class, which will randomly select 1 attribute to test its correlation with the label: 1 minutes Run the experimenter agent: Completed 14 random experiments in 16 seconds. (1.14 seconds per experiment)	This famous classification experiment tests INQLE's ability to test 4 different numeric attributes to predict the class of Iris plant, using a small data set.