miscalculation of scores

[alaineiturria]

To calculate the size of the windows, is it calculated considering the number of anomalies per data set present in the combined_labels.json file or those present in the combined_windows.json file?

I have been able to observe that the scores in the results/null/null_standard_scores.csv file are obtained using the combined_windows.json file, however, it seems to me that the anomalies present in combined_labels.json file are used to calculate the rest of the scores.

[subutai]

The window size calculation is described in the paper: "we define anomaly window length to be 10% the length of a data file, divided by the number of anomalies in the given file."

For each data file, the number of anomalies and their centers are given in the combined_labels.json file. The resulting windows are then put in combined_windows.json. This whole process is run by the combine_labels.py script.

Some more details are in the NAB whitepaper

For questions like these it might be better to ask on the HTM NAB Forum unless you see a specific bug we need to address.

Hope this helps!