I am looking at cleaning up the display of floats in the dashboard.
Currently:
For arguments, NumberEntryFloat will display 3+precision significant figures, possibly in scientific notation. The value is still rounded to an absolute precision before being submitted. This can lead to situations where the submitted value is different from the one displayed (due to the significant figures not capturing the absolute precision).
In datasets, all floating point values are displayed without scientific notation. This ensures that the displayed value accurately respects precision. This is inconsistent with (1).
One option would be to continue treating precision as the absolute rounding granularity and to introduce an optional sigfigs parameter in the experiment API to address the problem in (1).
Does this sound reasonable? Other thoughts on how best we display floating point values for your use case? Any feedback is appreciated.