Part of #437, and depends on the <series> work in Phase 1.
The first type whose series holds raw observations rather than one value per category. A box per series, side by side along the horizontal axis — which is exactly aes(x = group, y = value), so a series being the group is the standard reading rather than a Doenet-specific one.
The five-number summary
<summaryStatistics> already computes min, Q1, median, Q3 and max, and its FIVE_NUMBER_SUMMARY names that set. The quartiles must come from the same definition — math-expressions' quantileSeq, interpolated percentiles — so that a box plot and a <summaryStatistics> of the same data never disagree on the page. Extract the computation into a shared utility rather than reimplementing it.
Note what that means and what the docs must not claim: these are interpolated percentiles, not Tukey's hinges, and they differ on some sample sizes. SummaryStatistics.js already carries the comment explaining why the term "five-number summary" is used unattributed.
Drawing
<rectangle> from Q1 to Q3, filled from the series' styleNumber.<line> across the box at the median.<line> whiskers from the box to the furthest observation within 1.5 IQR of it, with a <line> cap at each end.<point> for each observation beyond the whiskers.
The 1.5 IQR rule is the default and the only rule implemented; if a whiskers attribute is wanted later (minMax, a multiplier), it slots in without changing the drawing.
Axis and labels
The horizontal axis carries one <tick-mark> per box, labeled with the series' label. That is the same categorical axis a bar chart draws, so categories is not used by this type — the boxes are the categories — and giving both should warn.
The vertical axis is the data's scale, so unlike a bar chart it does not have to include zero: bounds come from the observations and the outliers, rounded out to the next tick.
Public properties
Each <series> exposes its own summary — minimum, quartile1, median, quartile3, maximum, and the outliers — so a named series can be read into a sentence or an <answer>. This is the same argument #1833 made for binning in the worker rather than in scipy.
Tests
Quartiles agreeing with <summaryStatistics> on the same data, including the sample sizes where hinges and interpolated percentiles differ; whisker ends landing on real observations rather than on the fence; outliers drawn as points and not as whisker ends; a series with one observation and with none; several series side by side; the annotation tree.
🤖 Generated with Claude Code
https://claude.ai/code/session_01RqRJ3QoH4UrHFR41aAkN8e
Part of #437, and depends on the
<series>work in Phase 1.The first type whose series holds raw observations rather than one value per category. A box per series, side by side along the horizontal axis — which is exactly
aes(x = group, y = value), so a series being the group is the standard reading rather than a Doenet-specific one.The five-number summary
<summaryStatistics>already computes min, Q1, median, Q3 and max, and itsFIVE_NUMBER_SUMMARYnames that set. The quartiles must come from the same definition — math-expressions'quantileSeq, interpolated percentiles — so that a box plot and a<summaryStatistics>of the same data never disagree on the page. Extract the computation into a shared utility rather than reimplementing it.Note what that means and what the docs must not claim: these are interpolated percentiles, not Tukey's hinges, and they differ on some sample sizes.
SummaryStatistics.jsalready carries the comment explaining why the term "five-number summary" is used unattributed.Drawing
<rectangle>from Q1 to Q3, filled from the series'styleNumber.<line>across the box at the median.<line>whiskers from the box to the furthest observation within 1.5 IQR of it, with a<line>cap at each end.<point>for each observation beyond the whiskers.The 1.5 IQR rule is the default and the only rule implemented; if a
whiskersattribute is wanted later (minMax, a multiplier), it slots in without changing the drawing.Axis and labels
The horizontal axis carries one
<tick-mark>per box, labeled with the series'label. That is the same categorical axis a bar chart draws, socategoriesis not used by this type — the boxes are the categories — and giving both should warn.The vertical axis is the data's scale, so unlike a bar chart it does not have to include zero: bounds come from the observations and the outliers, rounded out to the next tick.
Public properties
Each
<series>exposes its own summary —minimum,quartile1,median,quartile3,maximum, and the outliers — so a named series can be read into a sentence or an<answer>. This is the same argument #1833 made for binning in the worker rather than in scipy.Tests
Quartiles agreeing with
<summaryStatistics>on the same data, including the sample sizes where hinges and interpolated percentiles differ; whisker ends landing on real observations rather than on the fence; outliers drawn as points and not as whisker ends; a series with one observation and with none; several series side by side; the annotation tree.🤖 Generated with Claude Code
https://claude.ai/code/session_01RqRJ3QoH4UrHFR41aAkN8e