Дмитрий
56 lines
2.3 KiB
JavaScript
56 lines
2.3 KiB
JavaScript
#!/usr/bin/env node
|
|
/**
|
|
* Accuracy runner — прогоняет 20 промптов через classifier (без LLM, regex only)
|
|
* и выдаёт отчёт «правильно/неправильно» по каждому пункту.
|
|
*
|
|
* Использовать перед регистрацией router-prehook в settings.json.
|
|
*/
|
|
|
|
import { readFileSync } from 'fs';
|
|
import { classifyByRegex } from './router-classifier-regex-fallback.mjs';
|
|
import { loadRegistry } from './registry-load.mjs';
|
|
|
|
function main() {
|
|
const promptsFile = process.argv[2] || 'tools/router-test-prompts.json';
|
|
const data = JSON.parse(readFileSync(promptsFile, 'utf-8'));
|
|
const registry = loadRegistry({ useCache: false });
|
|
|
|
let correctType = 0, correctNode = 0, correctMicro = 0, total = data.prompts.length;
|
|
const failures = [];
|
|
|
|
for (const p of data.prompts) {
|
|
const r = classifyByRegex(p.text, registry);
|
|
const typeOk = r.taskType === p.expectedType;
|
|
const nodeOk = r.recommendedNode === p.expectedNode;
|
|
const microOk = r.micro === p.expectedMicro;
|
|
if (typeOk) correctType++;
|
|
if (nodeOk) correctNode++;
|
|
if (microOk) correctMicro++;
|
|
if (!typeOk || !nodeOk || !microOk) {
|
|
failures.push({
|
|
prompt: p.text,
|
|
expected: { type: p.expectedType, node: p.expectedNode, micro: p.expectedMicro },
|
|
actual: { type: r.taskType, node: r.recommendedNode, micro: r.micro },
|
|
deltas: { type: !typeOk, node: !nodeOk, micro: !microOk },
|
|
});
|
|
}
|
|
}
|
|
|
|
console.log('=== Accuracy Report ===');
|
|
console.log(`Type accuracy: ${correctType}/${total} = ${(correctType / total * 100).toFixed(1)}%`);
|
|
console.log(`Node accuracy: ${correctNode}/${total} = ${(correctNode / total * 100).toFixed(1)}%`);
|
|
console.log(`Micro accuracy: ${correctMicro}/${total} = ${(correctMicro / total * 100).toFixed(1)}%`);
|
|
console.log('');
|
|
console.log(`Failures (${failures.length}):`);
|
|
for (const f of failures) {
|
|
console.log(` «${f.prompt}»`);
|
|
console.log(` expected: type=${f.expected.type}, node=${f.expected.node}, micro=${f.expected.micro}`);
|
|
console.log(` actual: type=${f.actual.type}, node=${f.actual.node}, micro=${f.actual.micro}`);
|
|
}
|
|
|
|
const passOverall = (correctType + correctNode + correctMicro) / (total * 3);
|
|
process.exit(passOverall >= 0.75 ? 0 : 1);
|
|
}
|
|
|
|
main();
|