Data pre processing with different models in NAB

Engineering NAB
1

Hi there,

I am working on comparisons between LSTM using different data reduction techniques and models results given by the NAB. As my focus is computing cost saving, I am interested to know more in detail process that was followed to get given models results (like numenta, randomCutForest and skyline) and especially if there was some pre-processing applied on the data before using these models. For example, using buckets encoding as quantification for HTM or time aggregation. If I’m not mistaken, I did not find any specific information about that in the NAB whitepaper. Is there any track somewhere of these process related to NAB results?

Thank you.
YC

1 Like
2

You mean like aggregation / standardization? I believe this is allowed, and different solutions can implement whatever preprocessing they like. You’ll have to look into the different detectors to see exactly what each one did, I think.

1 Like
3

Thank you for your quick reply.
Yes and any kind of data reduction. I have been looking at implementations just like you said and have found related information, thank you. Do NAB numenta results come after using time aggregation as the HTM studio ?