Adding Context to Training Data

Hi all,

I want to make an anomaly detector for insertion loss data (S21) over frequency. I have multiple cable types (varying in length, connector types, pitch, etc). Can I give this information to the detector or should I not? I have around 120,000 signals worth of data to train it on.

[dpb]

That would all depend upon what you expect your detector to require as inputs to use.

If it were to be simply a pass/fail logical output, you could try without any distinction and see what happens.

Preliminary exploring by scatter plots of the data by the various factors initially would never be a bad idea...

[Amber]

@dpb Thank you for your reply. It's good to know that it is possible.

And all of that information is encoded in the part number so it would be trivial to get that information to it in use.

I certainly can't argue with trying without the distinction to see what happens!

I fear I won't be able to give results for a while since I don't currently have the resources to test it, we are in the early stages of development on this.

[dpb]

However, I'd expect the real world will intrude and that different (classes of) parts are going to have different characteristics.