Does the data only have categorical inputs?

[Francis Duan]

I have noticed that most of the input entries are zero, and I am wondering if the attached dataset has any numerical inputs (sensor readings, etc.)? If there is, are you using them with one-hot encodings directly?

[Kyle Hundman]

Yes the telemetry channel's recent history is included. See the 'Data' section in the Readme for more details.

[Francis Duan]

Hi thanks for the reply. But my question was whether you are using numerical values and one-hot encoding values together?

Does it make sense to separate the two? Since numerical values might have very different latent information compared with categorical values

[Kyle Hundman]

Yes they are used together, this is shown here:

There's more discussion in my response to Q3 in #8 as well as in the paper.

It definitely could make sense to separate the two.

[Francis Duan]

Thanks a lot for this, sorry I did not realize the original y is also an input.

So the algorithm is transforming a Multivariate time series (MTS) into a univariate one (output y hat), with the target y included in the MTS as well. Is this correct?

[Kyle Hundman]

This is explained in the "Methods" section in the paper:

Prior values of y are inputs for the prediction of the current y value.

[Francis Duan]

ok thank you