Protodash Timeseries

[firmai]

In the HELOC example you metioned that Protodash can be applied to time series, but how would one go about doing that, do you perhaps have an example or can you offer some advice?

[sadhamanus]

We dont have a specific example, but the point there is that because of the internals of the method it would also be appropriate for such data. In that sense it would be used the same way where you have a data point that you want to explain and it would select from your training set similar points, where consecutive points may be correlated.

[firmai]

Thanks for the reply, but I don't think I will be happy with this answer until I see an example. Thanks again for brining this into the public. Hopefully someone can find a way to do this at some point.

[wendywangwwt]

I have the same question as @firmai . Given that many time series models, like ARIMA, only uses the time series data point with no X but Y, how can we apply protoDash on such model and data?

[sadhamanus]

So your question seems to be for univariate time series which is a bit different than the previous question which was for multivariate time series where you do have an X, as HELOC is that kind of dataset. For univariate where you just predict based on Y, exemplar based explanations dont make much sense. So use of protodash in this scenario is hard to imagine.