MOA is an open source framework for Big Data stream mining. It includes a collection of machine learning algorithms (classification, regression, clustering, outlier detection, concept drift detection and recommender systems) and tools for evaluation.
And the output goes as follows (only first three rows):
learning evaluation instances
evaluation time (cpu seconds)
model cost (RAM-Hours)
Exact Match
Accuracy
Hamming Score
Precision
Recall
F-Measure
868.0
0.509275225
0.0
0.9665513264129181
0.0
0.9665513264129181
0.0
0.0
0.0
1736.0
1.075152357
0.0
0.9648414985590779
0.0
0.9648414985590779
0.0
0.0
0.0
2604.0
1.48700487
0.0
0.9665770265078756
0.0
0.9665770265078756
0.0
0.0
0.0
From this output you can see some issues:
No values are returned for the metrics of Accuracy, Precision, Recall and F-Measure. All of them have a value of “0.0” for every window of samples evaluated.
Exact Match and Hamming Score have equals values.
The value for Exact match is too high, since it forces the prediction to have exactly the same labels as the testing sample.
This behavior seems to be present in every multi-label stream and multi-label model I tested.
I run the following command:
EvaluatePrequentialMultiLabel -l (multitarget.BasicMultiLabelClassifier -l multilabel.MultilabelHoeffdingTree) -s (generators.multilabel.MultilabelArffFileStream -l 20 -f /tmp/datasets/20ng/meka/20NG-F.arff) -f 868 -q 868 -w 868
And the output goes as follows (only first three rows):
From this output you can see some issues:
This behavior seems to be present in every multi-label stream and multi-label model I tested.
Thanks in advanced.