Edit on GitHub

starplast.strategy_panel

The Strategies tab: thirty-nine ways to infer something, each explained, runnable and self-testing.

Docked to the right of Evidence and Analysis. The top half lists the strategies by family, with the verdict each one earned when its self-test was run on the shipped data. The bottom half shows the selected one as a CARD first (strategy_card.StrategyCard): what it answers, the same four bars for every strategy (better than chance, reach, and two plain task metrics, each against chance), Run / Test / Details, four collapsed lines about its test, and one real failure beside one real success. Details ▸ opens the three tabs that hold everything else:

Guide     what it infers, why that works, how it fails, a step-by-step walkthrough, and how
          it is tested -- the text a person needs before believing its output
Settings  its parameters, each with a tooltip saying why it exists, and Run / Test / Stop
Results   the summary, the tables it produced, and the verdict of the last self-test

A strategy runs on the window's job runner, so it appears in the Jobs panel and can be stopped there like anything else. Results tables are registered through AnalysisPanel.results_table when the analysis panel is present, so they carry the same row action and right-click save as every other results table in the application.

Nothing here computes anything: strategies and strategy_catalog do, and this is their face.

DATA = '/home/runner/work/starplast/starplast/starplast/data'

MEASURED = '/home/runner/work/starplast/starplast/starplast/data/strategy_selftests.json'

TABLES = 6

VERDICT_COLORS = {'PASS': '#2e8b57', 'FAIL': '#c0392b', 'INCONCLUSIVE': '#888888'}

GRADE_COLORS = {'reliable': '#2e8b57', 'works when tuned': '#b8860b', 'weak': '#cc7a00', 'no skill': '#c0392b', 'untestable': '#888888'}

BUTTON_TIPS = {'run': 'Run the selected strategy with these settings on the whole table, in the background. Its tables appear under Results; a strategy that builds a map can then be shown on it.', 'test': "Run the strategy's self-test: hide information that is already known, ask the strategy for it back, and compare the answer with the same procedure on shuffled data. PASS means it beat that null by the stated margin on these settings.", 'stop': 'Ask the running strategy to stop. It stops at its next checkpoint, usually within a few seconds; nothing is written anywhere by a stopped run.', 'map': 'Put the map this result was computed on into the central view, colored by its clusters or modules, so the result can be looked at rather than only read.', 'genes_file': 'Read gene identifiers from a text or CSV file: every token in the first column, or the whole file if it is a plain list. Anything not found is reported.', 'genes_example': 'Fill the list with the members of one known category of the default label, so a gene-list strategy can be tried before you have a list of your own.', 'genes_gate': 'Use the genes currently gated on the map, so a structure you drew around can be handed straight to a gene-list strategy.', 'genes_test': 'Would the data have found these genes? Their labels are hidden all at once and five fast strategies are asked for them back; the answer opens in the evidence panel, gene by gene. Genes with no label are left out.', 'tuned': 'Fill the settings with the combination calibration found best for this strategy (chosen on some held-out seeds and confirmed on others). The held-out label or gene list stays yours to choose.', 'filter': "Type to show only strategies whose name, method, question or family contains the text -- for example 'list', 'network', 'HDBSCAN', 'logistic' or 'Plasmodium'.", 'details': 'Open everything the card leaves out: the Guide (method, techniques, walkthrough, every scorecard metric), the Settings (each parameter, the tuned setting, Stop) and the Results tables of the last run or test.', 'back': "Back to the strategy's card: the four bars, the test explained, and the worked examples."}

def load_measured(organism: str) -> dict:

The shipped self-test verdicts for one organism, or nothing if they were never measured.

def shipped_card(key: str, organism: str, measured: dict | None = None) -> dict:

metric -> (value, low, high) measured on the shipped data: the calibration's defaults with their 95% interval where the sweep recorded scorecards, else the single self-test.

A module function, not only a method, because the Start-here tab shows the same four bars for the strategies it recommends and must not compute them a second way.

def headline(key: str, organism: str, measured: dict | None = None) -> tuple:

(card, verdict, grade, basis) for one strategy's four bars, as StrategyCard takes them.

The calibration sweep's defaults with their 95% intervals where it measured the strategy on this space, else the single shipped self-test, else nothing and a line saying so.

class StrategyPanel(PyQt6.QtWidgets.QWidget):

The strategy list, a guide to the selected strategy, its settings, and its results.

StrategyPanel( nodes: pandas.DataFrame, runner=None, analysis=None, organism: str | None = None, graph=None, other=None, gated=None, parent=None)

Build the panel over one organism's table. Nothing is computed until something is run.

def embedding_ready(unknown):