Impute: allow user-defined value for "no data"

[wvdvegte]

What's your use case?
In Orange, cells with missing data are shown as "?" in Data Table. These are also the cells that are affected by impute. But in some standards, missing data is represented by some fixed value. For instance, in files produced by Garmin tracking devices, a missing heart rate is shown as "255" (apparently this is a heart rate never measured in real humans, also the max. value fitting into 8 bits).

What's your proposed solution?
Allow user specification of the value that corresponds to "no data"

Are there any alternative solutions?
Convert the "no data" to "?" using Formula, e.g., float (nan) if heart_rate == 255 else heart_rate - however this is not a very straightforward solution.

[markotoplak]

Did you try the Impute widget with its Fixed option?

[wvdvegte]

Yes, but that replaces the "?" values with a fixed value. It still needs an 'official' "?" on the input.

[markotoplak]

Oh, I misunderstood, OK, so you want the reverse of Impute. :)

[janezd]

I understand your frustration, I know this situation.

In general we avoid implementing data editing that can be done in Excel. This one is on the border.

We discussed different approaches, from the one that you mentioned (Formula) to changing Impute or defining the representation of the missing value in the File and CSV Importer. At the end we decided that the best place could be the Domain Editor because the presentation for the missing values is usually a property of a variable.

This is complicated to implement, though, so do not expect a quick solution...

[wvdvegte]

I'm not in a hurry - for now, the formula-based solution works fine and it is simple enough (though not obvious and therefore potentially user-unfriendly)