Simplify default cut labels
#422
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
`Statistics.quantile` returns values which are not the most appropriate to generate labels. It is more intuitive to choose values from the actual data, which are likely to have fewer decimals and make more sense for users. Unfortunately, since we use intervals closed on the left, we cannot use any of the seven standard definitions of quantiles. Type 1 is the closest, but we have to take the value next to it as a cutpoint to prevent it from being included into the next quantile group. This gives essentially consistent group attributions to R's `Hmisc::cut2` or `cut(x, quantile(x, (0:n)/n, type=1), include.lowest=T))`, though with different cutpoints in labels.
1) The quantile number isn't needed in most cases in the label, and anyway it's shown when printing an ordered `CategoricalValue`. Only use it by default when `allowempty=true` to avoid data-dependent errors if there are duplicate levels. 2) Round breaks by default to a number of significant digits chosen by `sigdigits`. This number is increased if necessary for breaks to remain unique. This generates labels which are not completely correct as rounding may make the left break greater than a value which is included in the interval, but this is generally minor and expected. Taking the floor rather than rounding would be more correct, but it can generate unexpected labels due to floating point trickiness (e.g. `floor(0.0003, sigdigits=4)` gives 0.0002999). This is what R does. Add a deprecation to avoid breaking custom `labels` functions which did not accept `sigdigits`.
andreasnoack
approved these changes
May 17, 2025
|
I think this looks good but I can see that it relies on functionality that isn't supported in 1.6. With a new LTS, I'm fine with increasing the minimum version but maybe it is easy to adjust the implementation to avoid the newer functionality. It also looks like there is still a segfault on Windows. |
|
Yeah, I think I can easily use a slightly less efficient approach on Julia 1.6 (generating a format string on each call). The segfault on 32-bit Windows is always the same one related to Plots. I'll have to report it to them. While you're at it, if you can have a look at the latest version of #416. :-) |
bkamins
approved these changes
May 18, 2025
andreasnoack
force-pushed
the
nl/quantilecut
branch
from
e5d84c7 to
a33a853
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The quantile number isn't needed in most cases in the label, and anyway it's shown when printing an ordered
CategoricalValue. Only use it by default whenallowempty=trueto avoid data-dependent errors if there are duplicate levels.Round breaks by default to a number of significant digits chosen by
sigdigits. This number is increased if necessary for breaks to remain unique. This generates labels which are not completely correct as rounding may make the left break greater than a value which is included in the interval, but this is generally minor and expected. Taking the floor rather than rounding would be more correct, but it can generate unexpected labels due to floating point trickiness (e.g.floor(0.0003, sigdigits=4)gives 0.0002999). This is what R does.Add a deprecation to avoid breaking custom
labelsfunctions which did not acceptsigdigits.Fixes #381. This is on top of #416 for simplicity but essential orthogonal.
Some points worth discussing:
@sprintf, which apparently matches the C standard. We could use 4 significant digits by default to fix this, but it would make the output more verbose than it needs to be in general. We could also call@sprintfwith both%fand%eand chose the shortest but it's a bit ad-hoc. I couldn't find a "better" printing method in packages (e.g. Format.jl).string, so they never use scientific notation. This fixes the previous problem in most cases, but if you have very large values it could be annoying. Not sure.