matlab/matlab-agentic-toolkit / skills-catalog/ai-and-statistics/matlab-engineer-tabular-features
matlab-engineer-tabular-features
UnverifiedUse when engineering or selecting the best features for single-response classification or regression in MATLAB, whatever the data's modality — for non-tabular data it routes extraction to a domain skill, then selects, assesses, and delivers on the resulting table. Not for multi-response problems, model training, or raw data acquisition.
scripts/plotSelectionStability.m
View on GitHubfunction Fig = plotSelectionStability(VoteTable, NSelect, Stability, OutputDir, Options)
%plotSelectionStability Stability of the selection decision: per-feature re-selection frequency.
%
% Fig = plotSelectionStability(VoteTable, NSelect, Stability, OutputDir) writes
% fig_selection_stability.svg -- the candidate features in consensus-RANK order
% (delivered set at the top) as horizontal bars whose length is how often each
% feature was re-selected when the training rows were resampled. This audits the
% decision drawn by the companion plotSelectionDecision (fig_selection_decision):
% same features, same rank order, so the two figures line up.
%
% A horizontal line marks the cut (NSelect); a vertical line marks the frequency
% floor (reference only -- it gates nothing).
%
% Two stability reads, two encodings:
% * With the fixed-pool read alone (Stability), one bar series is drawn;
% delivered features are blue, the rest grey.
% * When a generation-stability read is also supplied
% (Options.GenStabSelection), GROUPED bars are drawn -- fixed pool (blue)
% beside generation (orange) -- so the per-feature gap is the generation
% variance. A feature present in only some subsamples' pools DRIFTED: it has
% no generation bar and its row is marked with a red band.
%
% Reading the two reference lines together is the whole point:
% * a SHORT bar left of the floor (shipped) is a FRAGILE pick -- the consensus
% kept it, resampling rarely did;
% * a LONG bar below the cut (not shipped) is a ROBUST NEAR-MISS -- resampling
% liked it more than the consensus did.
% Position encodes what shipped; length encodes how stable that was.
%
% With no stability read (Stability = []), nothing is drawn and the function
% returns [] -- the decision itself is always available from plotSelectionDecision.
%
% Inputs:
% VoteTable - table(Feature, Score, Rank, Selected) from runConsensusSelection
% (consensus order); Rank orders the bars, Selected marks the
% delivered set
% NSelect - (1,1) double, the number of features the cut kept (the cut line)
% Stability - the fixed-pool read: a struct with SelectionFrequency (table
% Feature, Frequency), Nogueira (1,1), NSelectPerResample (1,:);
% or [] to draw nothing
% OutputDir - (1,1) string, destination folder for the SVG
% Options.GenStabSelection - the generation-stability read
% (assessGenerationStability): a struct with SelectionFrequency
% (over the intersection universe) and, optionally, Drift. When
% present, adds the generation series and the drift band.
% Default [] (fixed pool only).
% Options.FrequencyFloor - (1,1) double in [0,1], the resample-frequency
% reference line (default 0.5). DIAGNOSTIC only; it references the
% shipped fixed-pool read and changes nothing.
%
% Output:
% Fig - the figure handle (written to fig_selection_stability.svg), or [] when
% no stability read was supplied
% Copyright 2026 The MathWorks, Inc.
arguments
VoteTable table
NSelect (1,1) double {mustBePositive}
Stability
OutputDir (1,1) string
Options.GenStabSelection = []
Options.FrequencyFloor (1,1) double {mustBeGreaterThanOrEqual(Options.FrequencyFloor, 0), ...
mustBeLessThanOrEqual(Options.FrequencyFloor, 1)} = 0.5
end
Floor = Options.FrequencyFloor;
CutClr = [0.15 0.15 0.15]; % near-black: distinct from the orange generation bars
BlueClr = [0.20 0.50 0.80];
OrangeClr = [0.90 0.55 0.10];
GreyClr = [0.72 0.72 0.72];
HasStability = ~isempty(Stability) && isstruct(Stability) ...
&& isfield(Stability, "SelectionFrequency") && istable(Stability.SelectionFrequency);
GS = Options.GenStabSelection;
HasGenStab = ~isempty(GS) && isstruct(GS) ...
&& isfield(GS, "SelectionFrequency") && istable(GS.SelectionFrequency);
if ~HasStability
% No stability read -- the decision lives in plotSelectionDecision; nothing
% to draw here.
Fig = [];
return;
end
% One full-width panel + a how-to-read caption strip along the bottom.
Fig = figure(Name="Selection Stability", Units="centimeters", ...
Position=[2 2 18 17], Visible="off", Color="w");
Ax = axes(Fig, OuterPosition=[0 0.17 1 0.83]); % leave room for the caption
paintStability(Ax, VoteTable, NSelect, Stability, GS, HasGenStab, ...
Floor, CutClr, BlueClr, OrangeClr, GreyClr);
subtitle(Ax, stabilitySubtitle(HasGenStab));
drawCaption(Fig, HasGenStab);
exportgraphics(Fig, fullfile(OutputDir, "fig_selection_stability.svg"), ...
ContentType="vector");
end
function paintStability(ax, VoteTable, NSelect, Stability, GS, HasGenStab, Floor, ...
CutClr, BlueClr, OrangeClr, GreyClr)
%paintStability Per-feature selection frequency in consensus-rank order.
% One bar series for the fixed-pool read; grouped fixed-pool/generation bars
% when a generation read is supplied, with drifted features (absent from the
% generation intersection universe) marked by a red band.
VoteSorted = sortrows(VoteTable, "Rank", "ascend");
Names = string(VoteSorted.Feature);
FreqFixed = freqByName(Names, Stability.SelectionFrequency);
% One labelled bar per feature does not scale, so cap the count -- but never
% below the cut, so every delivered bar is always shown.
MaxBars = max(25, NSelect);
nShow = min(numel(Names), MaxBars);
ShowNames = Names(1:nShow);
ShowFixed = FreqFixed(1:nShow);
IsDelivered = logical(VoteSorted.Selected(1:nShow));
y = 1:nShow;
if HasGenStab
% Generation frequency is defined only over the intersection universe;
% a feature absent from it (NaN here) DRIFTED across subsamples.
FreqGS = freqByName(ShowNames, GS.SelectionFrequency);
Drift = isnan(FreqGS);
GSHeights = FreqGS;
GSHeights(Drift) = 0; % keep the group slot; mark via band
b = barh(ax, y, [ShowFixed GSHeights], 0.82, FaceColor="flat");
b(1).FaceColor = BlueClr;
b(2).FaceColor = OrangeClr;
hold(ax, "on");
for i = find(Drift(:))'
patch(ax, [0 1 1 0], [y(i)-0.5 y(i)-0.5 y(i)+0.5 y(i)+0.5], [1 0.85 0.85], ...
FaceAlpha=0.55, EdgeColor=[0.7 0 0], LineStyle="--");
end
else
% Single series: colour encodes delivered (blue) vs not (grey).
b = barh(ax, y, ShowFixed, FaceColor="flat");
b.CData = repmat(GreyClr, nShow, 1);
b.CData(IsDelivered, :) = repmat(BlueClr, nnz(IsDelivered), 1);
hold(ax, "on");
end
if NSelect < nShow
yline(ax, NSelect + 0.5, "--", sprintf("cut = %d", NSelect), Color=CutClr, ...
LineWidth=1.3, LabelHorizontalAlignment="right", Interpreter="none");
end
xline(ax, Floor, ":", [sprintf("floor = %.2f", Floor); "(fixed pool)"], ...
Color=[0.5 0 0.5], LineWidth=1.3, LabelVerticalAlignment="bottom", ...
Interpreter="none");
hold(ax, "off");
set(ax, YDir="reverse", YTick=y, YTickLabel=ShowNames, TickLabelInterpreter="none");
% Shrink the y-tick font when the bars are many or the names long, so long
% engineered names stay legible instead of overrunning their neighbours.
LongestName = max([strlength(ShowNames(:)); 1]);
YFontSize = min(11, max(7, round(min(200 / nShow, 320 / LongestName))));
ax.YAxis.FontSize = YFontSize;
xlim(ax, [0 1]);
xlabel(ax, "Selection frequency across resamples");
NOmitted = numel(Names) - nShow;
if NOmitted > 0
title(ax, sprintf("Selection Stability (top %d; %d more omitted)", nShow, NOmitted));
else
title(ax, "Selection Stability");
end
if HasGenStab
legend(ax, b, ["fixed pool", "generation"], Location="southoutside", ...
Orientation="horizontal", Box="off");
end
grid(ax, "on");
end
function F = freqByName(Names, FreqTable)
%freqByName Selection frequency aligned to Names; NaN where a name is absent from
% the table (used to detect drift on the generation read).
[Present, Loc] = ismember(Names, string(FreqTable.Feature));
F = nan(numel(Names), 1);
F(Present) = FreqTable.Frequency(Loc(Present));
end
function S = stabilitySubtitle(HasGenStab)
%stabilitySubtitle One-line legend under the main title, per mode.
if HasGenStab
S = "blue = fixed pool | orange = generation | left of floor = fragile | red band = drift";
else
S = "blue = shipped | grey = not shipped | left of floor = fragile";
end
end
function drawCaption(Fig, HasGenStab)
%drawCaption Baked-in how-to-read guide along the bottom strip. Static encoding
% only -- run-specific reads (Nogueira value, named fragile picks, the
% generation-variance gap) stay in the report prose.
if HasGenStab
Caption = [ ...
"How to read — this audits the selection decision (the consensus-score elbow figure) under resampling; features are in the same consensus-rank order, shipped set at the top."; ...
"Bar length is how often a feature was re-selected under resampling — a short bar left of the floor is a fragile pick, a long bar below the cut is a robust near-miss, and a red band marks a drifted feature present in only some subsamples' pools."; ...
"The floor is a reference only and gates nothing."];
else
Caption = [ ...
"How to read — this audits the selection decision (the consensus-score elbow figure) under resampling; features are in the same consensus-rank order, shipped set at the top."; ...
"Bar length is how often a feature was re-selected under resampling — a short bar left of the floor is a fragile pick, and a long bar below the cut is a robust near-miss."; ...
"The floor is a reference only and gates nothing."];
end
annotation(Fig, "textbox", [0.02 0.01 0.96 0.15], String=Caption, ...
EdgeColor=[0.8 0.8 0.8], BackgroundColor=[0.97 0.97 0.97], ...
FontSize=10, VerticalAlignment="middle", HorizontalAlignment="left", ...
Interpreter="none");
end
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 466k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |