SPB Git

Repository Public

ChiSquareTools.swift 12.1 KB 179 lines swift

1import Foundation2import SheetCore3import Statistics4import AnalysisKit56/// Chi-square goodness-of-fit test.7enum ChiSquareGoodnessOfFitTool {8    static let descriptor: ToolDescriptor = {9        var t = ToolDescriptor(10            id: "compare.chisq-gof", title: L("Chi-Square Goodness of Fit"), group: Common.groupCompare, systemImage: "checklist",11            summary: L("Tests whether observed category counts follow hypothesised proportions."),12            fields: [13                .range("data", L("Data range")),14                .column("observed", L("Observed counts")),15                .column("labels", L("Category labels"), numeric: false, optional: true),16                .choice("expectedKind", L("Expected distribution"), [FieldOption("equal", L("Equal proportions")), FieldOption("counts", L("Expected counts column")),17                                                                    FieldOption("probabilities", L("Probabilities column"))]),18                .column("expected", L("Expected column"), optional: true, visible: { $0.choice("expectedKind") != "equal" }),19                .integer("estimated", L("Parameters estimated from the data"), 0, min: 0, max: 50,20                         help: L("Reduces the degrees of freedom: df = categories − 1 − estimated parameters.")),21                .alpha(L("Significance level α")),22            ],23            info: [24                L("Pearson statistic Σ(O − E)²/E; the p-value is the upper tail of χ² with k − 1 − m degrees of freedom."),25                L("Expected counts must have the same total as the observed counts; probabilities are normalised to sum to 1."),26                L("The χ² approximation needs expected counts that are large enough (rule of thumb: all ≥ 5)."),27            ],28            run: run)29        t.demo = { _, v in v.set("data", range: "Groups!H1:K4"); v.set("observed", column: 1); v.set("labels", column: 0) }30        return t31    }()3233    static func run(_ input: ToolInput) throws -> ToolResult {34        let v = input.values35        let alpha = try Common.alpha(v)36        let frame = try input.frame()37        let obsCol = try input.column("observed")38        let kind = v.choice("expectedKind")39        let expCol = kind == "equal" ? nil : input.optionalColumn("expected")40        if kind != "equal", expCol == nil { throw ToolError(L("Choose the column holding the expected counts or probabilities.")) }41        let labelCol = input.optionalColumn("labels")42        var series = [obsCol.numbers]43        if let expCol { series.append(expCol.numbers) }44        let (rows, kept) = DataFrame.listwise(series)45        guard rows.count >= 2 else { throw ToolError(L("At least two categories with numeric counts are required.")) }46        let observed = kept[0]47        let test: HypothesisTest48        let total = observed.reduce(0, +)49        let expected: [Double]50        switch kind {51        case "counts":52            expected = kept[1]53            test = try Inference.chiSquareGoodnessOfFit(observed: observed, expected: expected, estimatedParameters: v.int("estimated"))54        case "probabilities":55            let sum = kept[1].reduce(0, +)56            expected = kept[1].map { $0 / sum * total }57            test = try Inference.chiSquareGoodnessOfFit(observed: observed, probabilities: kept[1], estimatedParameters: v.int("estimated"))58        default:59            expected = Array(repeating: total / Double(observed.count), count: observed.count)60            test = try Inference.chiSquareGoodnessOfFit(observed: observed, estimatedParameters: v.int("estimated"))61        }62        let labels = rows.map { r in labelCol.map { $0.labels[r] }.flatMap { $0.isEmpty ? nil : $0 } ?? L("Category %d", r + 1) }63        var table: [[Value]] = [[.text(L("Category")), .text(L("Observed")), .text(L("Expected")), .text(L("Residual (O − E)/√E")), .text(L("Contribution to χ²"))]]64        for i in observed.indices {65            let d = observed[i] - expected[i]66            table.append([.text(labels[i]), .number(observed[i]), .finite(expected[i]), .finite(d / expected[i].squareRoot()), .finite(d * d / expected[i])])67        }68        table.append([.text(L("Total")), .number(total), .finite(expected.reduce(0, +)), .blank, .finite(test.statistic)])69        var extra: [ResultBlock] = [70            .table(ReportTable(title: L("Observed and expected counts"), rows: table, columnFormats: [2: .fixed(2), 3: .fixed(4), 4: .fixed(4)])),71            .chart(ChartBlock(title: L("Observed and expected counts"),72                              spec: ChartKit.columns(categories: labels, series: [(L("Observed"), observed), (L("Expected"), expected)], xTitle: L("Category"), yTitle: L("Count")),73                              summary: L("Grouped bars of observed and expected counts for %d categories.", observed.count), height: 220)),74        ]75        if let m = expected.min(), m < 5 {76            extra.insert(.warning(L("The smallest expected count is %@ (< 5): the χ² approximation may be unreliable.", Fmt.number(m, locale: input.locale))), at: 0)77        }78        return Common.testResult(test, title: L("Chi-Square Goodness of Fit"), subtitle: frame.label + " — " + obsCol.name, alpha: alpha,79                                 null: L("the observed counts follow the expected distribution"), locale: input.locale, extra: extra, sheetName: L("Goodness of fit"))80    }81}8283/// Chi-square test of independence from a contingency table or from two categorical columns.84enum ChiSquareIndependenceTool {85    static let descriptor: ToolDescriptor = {86        var t = ToolDescriptor(87            id: "compare.chisq-independence", title: L("Chi-Square Independence"), group: Common.groupCompare, systemImage: "tablecells",88            summary: L("Tests whether two categorical variables are independent, from a table of counts or from raw observations."),89            fields: [90                .choice("source", L("Input"), [FieldOption("table", L("Contingency table of counts")), FieldOption("columns", L("Two categorical columns"))]),91                .range("data", L("Data range"), help: nil),92                .column("rowFactor", L("Row variable"), numeric: false, visible: { $0.choice("source") == "columns" }),93                .column("colFactor", L("Column variable"), numeric: false, visible: { $0.choice("source") == "columns" }),94                .toggle("yates", L("Yates continuity correction (2 × 2 tables only)"), false),95                .alpha(L("Significance level α")),96            ],97            info: [98                L("Contingency table: the header row gives the column labels; a first column of text gives the row labels; every numeric column is a column of counts."),99                L("Two categorical columns: the observations are cross-tabulated; rows with a blank value in either column are dropped."),100                L("Expected counts are row total × column total / grand total; df = (r − 1)(c − 1). Cramér's V is computed from the uncorrected χ²."),101                L("Fisher's exact test is not available; with small expected counts interpret the result with caution."),102            ],103            run: run)104        t.demo = { _, v in v.set("data", range: "Groups!H1:K4") }105        return t106    }()107108    /// Cross-tabulates two label columns (levels in order of first appearance).109    static func crossTab(_ a: [String], _ b: [String]) -> (rows: [String], cols: [String], counts: [[Double]], dropped: Int) {110        var rows = [String](), cols = [String](), rowIndex = [String: Int](), colIndex = [String: Int]()111        var cells = [[Int]: Double]()112        var dropped = 0113        for (x, y) in zip(a, b) {114            guard !x.isEmpty, !y.isEmpty else { dropped += 1; continue }115            if rowIndex[x] == nil { rowIndex[x] = rows.count; rows.append(x) }116            if colIndex[y] == nil { colIndex[y] = cols.count; cols.append(y) }117            cells[[rowIndex[x]!, colIndex[y]!], default: 0] += 1118        }119        let counts = rows.indices.map { i in cols.indices.map { cells[[i, $0]] ?? 0 } }120        return (rows, cols, counts, dropped)121    }122123    static func run(_ input: ToolInput) throws -> ToolResult {124        let v = input.values125        let alpha = try Common.alpha(v)126        let frame = try input.frame()127        var rowLabels: [String], colLabels: [String], counts: [[Double]]128        var notes: [String?] = []129        var subtitle = frame.label130        if v.choice("source") == "columns" {131            let a = try input.column("rowFactor"), b = try input.column("colFactor")132            guard a.index != b.index else { throw ToolError(L("Choose two different columns.")) }133            let tab = crossTab(a.labels, b.labels)134            (rowLabels, colLabels, counts) = (tab.rows, tab.cols, tab.counts)135            if tab.dropped > 0 { notes.append(L("%d rows with a blank value were dropped.", tab.dropped)) }136            subtitle += " — " + a.name + " × " + b.name137        } else {138            let numeric = frame.columns.filter(\.isNumeric)139            guard numeric.count >= 2 else { throw ToolError(L("The contingency table needs at least two numeric columns of counts.")) }140            let labelCol = frame.columns.first { !$0.isNumeric && $0.textCount > 0 }141            let (rows, cols) = DataFrame.listwise(numeric.map(\.numbers))142            if rows.count < frame.rowCount { notes.append(L("%d rows with a missing count were dropped.", frame.rowCount - rows.count)) }143            colLabels = numeric.map(\.name)144            rowLabels = rows.map { r in labelCol.map { $0.labels[r] }.flatMap { $0.isEmpty ? nil : $0 } ?? L("Row %d", r + 1) }145            counts = rows.indices.map { i in cols.map { $0[i] } }146        }147        guard rowLabels.count >= 2, colLabels.count >= 2 else { throw ToolError(L("The table needs at least two rows and two columns.")) }148        let result = try Inference.chiSquareIndependence(counts, yatesCorrection: v.bool("yates"))149        func grid(_ title: String, _ m: Matrix, totals: Bool, format: CellFormat) -> ReportTable {150            var rows: [[Value]] = [[.text("")] + colLabels.map { .text($0) } + (totals ? [.text(L("Total"))] : [])]151            for i in rowLabels.indices {152                rows.append([.text(rowLabels[i])] + colLabels.indices.map { .finite(m[i, $0]) } + (totals ? [.finite(result.rowTotals[i])] : []))153            }154            if totals { rows.append([.text(L("Total"))] + result.columnTotals.map { .finite($0) } + [.finite(result.total)]) }155            var t = ReportTable(title: title, rows: rows)156            for c in 1...(colLabels.count + 1) { t.columnFormats[c] = format }157            return t158        }159        var extra: [ResultBlock] = []160        if result.minimumExpected < 5 {161            extra.append(.warning(L("The smallest expected count is %@ (< 5): the χ² approximation may be unreliable.", Fmt.number(result.minimumExpected, locale: input.locale))))162        }163        if v.bool("yates"), !(rowLabels.count == 2 && colLabels.count == 2) { notes.append(L("The Yates correction applies to 2 × 2 tables only and was not used.")) }164        extra += [165            .table(grid(L("Observed counts"), result.observed, totals: true, format: .integer)),166            .table(grid(L("Expected counts under independence"), result.expected, totals: false, format: .fixed(2))),167            .table(grid(L("Pearson residuals (O − E)/√E"), result.residuals, totals: false, format: .fixed(3))),168            .chart(ChartBlock(title: L("Observed counts"),169                              spec: ChartKit.columns(categories: rowLabels, series: colLabels.indices.map { j in (colLabels[j], rowLabels.indices.map { result.observed[$0, j] }) },170                                                     xTitle: "", yTitle: L("Count")),171                              summary: L("Grouped bars of the observed counts: %d rows by %d columns, %@ observations.", rowLabels.count, colLabels.count,172                                         Fmt.number(result.total, locale: input.locale)), height: 220)),173        ]174        return Common.testResult(result.test, title: L("Chi-Square Test of Independence"), subtitle: subtitle, alpha: alpha,175                                 null: L("the row and column variables are independent"), locale: input.locale, notes: notes, extra: extra,176                                 sheetName: L("Independence"))177    }178}179
Files main
Menu

Theme