DTW prediction - Distance Matrix Values are INF!

[SimonAmpleman]

Hello,
I am trying to use DTW with 6 floats per samples. I loaded 3 samples of around 200 events each for class 1 and about the same for class 2.

When I try to use the predict method each frame to get a realtime prediction, I get this error message :

GRT\ClassificationModules\DTW\DTW.cpp GRT::DTW::computeDistance:626 Distance Matrix Values are INF!

It shows on the console from time to time, maybe around each two seconds.

And for the prediction itself, it looks like random results. It switch between class 1 and 2 with large getMaximumLikelihood() such as "8".

I don't see anything wrong in my training data :

TrainingData.txt

I am on Windows platform, VS2017 with the latest master git source.

Thank you !

[nickgillian]

Hi Simon,

Your training data looks valid. It sounds like there might be some issues with the real-time data you where inputting to the trained model. Did you plot or look at the real-time data to see if there were any issues (such as NaNs in the data, or all zeros, or something like that)?