Representing -inf and inf in json

[oeway]

Currently the data range representation in yaml can support float infinity, but when trying to convert to json, since there is no native support for -inf and inf, we need to either convert to a string Infinity and -Infinity or set it to null which represent no boundary. Not sure if we need to support in the spec, but maybe as a discussion for future versions.

[FynnBe]

This can be done via pydantic: https://docs.pydantic.dev/latest/api/config/#pydantic.config.ConfigDict.ser_json_inf_nan
We could add an analog to https://bioimage-io.github.io/spec-bioimage-io/bioimageio/spec.html#dump_description for json, if there are use cases??

Accepting "inf" is not a problem as pydantic used the builtin float() function (https://docs.pydantic.dev/latest/api/standard_library_types/#float), so "inf" is already valid (in a bioimage.yaml file, too).

[FynnBe]

The default value of null is actually a bit odd as it maps +inf and -inf to the same value...
So is there partial support for the constants somehow??
stings would fix it then I guess... reading the json as yaml and then parsing with our pydantic models will be compatible:

>>> float("Infinity")
inf
>>> float("-Infinity")
-inf
>>>

Should we change the ser_json_inf_nan config value to strings then, before someone relies on it being 'null'?

I can update the type annotation as dump_description really already returns a JsonValue (YAML just also allows for datetime objects, but when dumping timestamps are already serialized to string. Just when loading a YAML file the YAML parser already returns datetime objects.

[oeway]

Right, I think we can go for string representation.

Using null is not entirely nonsense,iIf we go for that I would interpret it differently, it means unbounded. Inf is not very standard as well, e.g. python has None type, but inf is should be write as float("inf") or import math;math.inf.