mariolpantunes / ml

A standard machine learning library written in Java. It contains several pieces of code/methods/functions that are related with ML and useful for my research. More functions and algorithms are being developed (mostly when I need them).

Hi there 👋

Mário antunes

Researcher | Machine Learning and Artificial Intelligence
Instituto de Telecomunicações, Aveiro, Portugal | TN-AV
University of Aveiro
[email protected] | [email protected]
ResearchGate | ORCID | Linkedin

ml's Issues

bug

The following code results in an Exception:
final double[] X = new double[] { 1, 2, 3 };
final double[] Y = new double[] { 0, 0.1, 0.1 };
new DFDT().elbow(X, Y);

bug BaseCurvature::knee and BaseCurvature::elbow invalid

Bug reproduction code:
final double[] X = new double[] { 1, 2, 3 };
final double[] Y = new double[] { 0, 0, 0 };
new DSDT().elbow(X, Y);
new DFDT().elbow(X, Y);
new DSDT().knee(X, Y);
new DFDT().knee(X, Y);

Explanation:
When all values of Y are the same (in the above example all are 0.0), the function BaseCurvature::knee (and BaseCurvature::elbow) will calculate the slope -- which would be 0.
However, the code only checks for:
if (slope > 0) {...} else {...}

Thus, the case of slope==0 falls into the else, and calls find_elbow(x, y), which in turn calls itRefinement(x, y), which in turn calls dsdt(..).
However, dsdt(..) calls:
m[] = ArrayUtils.csd(...)
which returns an empty array.
This then calls isoData(..) and cause
double t = mean(...)
to be t=NaN

I suggest fixing the bug in both the elbow(..) and knee(..) level, i.e., after the slope calculation add:
if (slope == 0) { // all values are the same
return rv; // -1
} else if (slope < 0) { ...}
} else { ... }