semantic-ds-toolkit/dist/src/tools/drift-eval.js

Performance-first semantic layer for modern data stacks - Stable Column Anchors & intelligent inference

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
const drift_1 = require("../drift");
function nowIso() { return new Date().toISOString(); }
function synthNormal(n, mean = 0, std = 1) {
    const out = [];
    for (let i = 0; i < n; i++) {
        const u = Math.random();
        const v = Math.random();
        const z = Math.sqrt(-2 * Math.log(u)) * Math.cos(2 * Math.PI * v);
        out.push(mean + std * z);
    }
    return out;
}
function fpFrom(values) {
    const min = Math.min(...values);
    const max = Math.max(...values);
    const uniq = new Set(values);
    return {
        dtype: 'float64',
        cardinality: uniq.size,
        regex_patterns: [],
        null_ratio: 0,
        unique_ratio: uniq.size / values.length,
        sample_values: values.slice(0, 20).map(String),
        min, max,
    };
}
function anchorFor(name, fp) {
    const ts = nowIso();
    return {
        dataset: 'eval',
        column_name: name,
        anchor_id: `eval:${name}`,
        fingerprint: JSON.stringify(fp),
        first_seen: ts,
        last_seen: ts,
        confidence: 0.95,
    };
}
async function evalNumericTrials(trials = 50) {
    const detector = new drift_1.DriftDetector({ enable_performance_mode: true });
    let fpCount = 0; // false positive
    let tpCount = 0; // true positive
    // False positives: same distribution
    for (let i = 0; i < trials; i++) {
        const base = synthNormal(20000, 100, 15);
        const cur = synthNormal(20000, 100, 15);
        const baseFp = fpFrom(base);
        const curFp = fpFrom(cur);
        const anchor = anchorFor('stable_metric', baseFp);
        const col = { name: 'stable_metric', values: cur, data_type: 'float64' };
        const res = await detector.detectDrift(anchor, col, curFp);
        if (res.drift_detected)
            fpCount++;
    }
    // True positives: significant mean shift
    for (let i = 0; i < trials; i++) {
        const base = synthNormal(20000, 50, 10);
        const cur = synthNormal(20000, 200, 10);
        const baseFp = fpFrom(base);
        const curFp = fpFrom(cur);
        const anchor = anchorFor('mean_shift_metric', baseFp);
        const col = { name: 'mean_shift_metric', values: cur, data_type: 'float64' };
        const res = await detector.detectDrift(anchor, col, curFp);
        if (res.drift_detected)
            tpCount++;
    }
    return {
        trials,
        false_positive_rate: fpCount / trials,
        true_positive_rate: tpCount / trials,
    };
}
async function main() {
    const stats = new drift_1.StatisticalTests();
    // Quick check of stats utils
    const ks = stats.kolmogorovSmirnovTest([1, 2, 3, 4], [10, 11, 12, 13]);
    console.log('KS sanity:', { statistic: ks.statistic.toFixed(4), p: ks.p_value.toFixed(4) });
    const numeric = await evalNumericTrials(30);
    console.log('Drift eval summary:', numeric);
}
main().catch(err => {
    console.error('drift-eval error', err);
    process.exit(1);
});
//# sourceMappingURL=drift-eval.js.map