q param in quantiles

[soxofaan]

openeo-processes/quantiles.json

Lines 36 to 41 in 60fb5e2

	"name": "q",
	"description": "Intervals to calculate quantiles for. Calculates q-quantiles with (nearly) equal-sized intervals.",
	"schema": {
	"type": "integer",
	"minimum": 2
	},

I just noticed we have this q parameter in quantiles.
A couple of remarks:

• this q is a pretty short short name and not very self-explanatory. We probably refer to the statistical concept of "q-quantiles", but numpy,quantile has a q param and that one is equivalent to the probabilities we already have, so that might be confusing
• " (nearly) equal-sized intervals" : why "nearly"? It's confusing to specify it like that at the moment. If you talk about cutting a "probability distribution" in "intervals", they can be exactly the same size I think. If we would talk about splitting a set of observations in subsets then it makes more sense to state they are nearly same size.
• "q: intervals to calculate quantiles for": I would add "number of intervals" for clarity

[soxofaan]

On a more general note: I find the probabilistic definition of this process a bit strange (parameter probabilities, talking about "probability distribution"). We don't work here with random/unknown variables, it's all fully defined observation data.
For example: https://numpy.org/doc/stable/reference/generated/numpy.quantile.html does not mention something like "probability"

[m-mohr]

Hmm, this process was one of the first ones I defined and nowadays I'd actually split this process into two processes and then, I think, most issues could be solved a bit easier. To avoid breaking changes we can only improve the documentation, but can't really change the names. The naming for statistical processes is more aligned with R, that's why this time it's a bit more confusing for Python users it seems. Thanks for the suggestions, I'll try to come up with a reasonable PR.

[m-mohr]

• " (nearly) equal-sized intervals" : why "nearly"? It's confusing to specify it like that at the moment. If you talk about cutting a "probability distribution" in "intervals", they can be exactly the same size I think.

I can't recall why this is mentioned here, I've tried to consult my sources when writing the process. I guess one of the implementations I've looked at mentioned it so I left it in brackets to get a bigger coverage of tools that can implement the process, but I can't find it anymore and it is indeed not helpful for implementations to leave this ambiguous. So I'll remove this.

[m-mohr]

• this q is a pretty short short name and not very self-explanatory. We probably refer to the statistical concept of "q-quantiles", but numpy,quantile has a q param and that one is equivalent to the probabilities we already have, so that might be confusing

I've opened a separate issue as this is breaking by itself, but we could work around it for now. For details see #293 and let me know there whether the non-breaking behavior should be implemented already in 1.x

[m-mohr]

PR is available for review: #294

[m-mohr]

All issues have been solved or are now part of a separate issue.