Datasets:
Download by-class.mjs from Talcoe/perch-trapset: direct link, hf CLI and curl.
- Browser
- Download file 2.81 kB
-
https://huggingface.co/datasets/Talcoe/perch-trapset/resolve/main/by-class.mjs
- Command line
-
hf download hf://datasets/Talcoe/perch-trapset/by-class.mjs
-
curl -L -o by-class.mjs https://huggingface.co/datasets/Talcoe/perch-trapset/resolve/main/by-class.mjs
2.81 kB
| // Breaks the ledger out by drug class: for each class in drug-classes.json, how many dictated | |
| // drug instances each output recovered and which ones it missed. It reads the per-clip table in | |
| // ledger.md, so the counts are the ledger's own, regrouped; nothing is rescored. Move a drug to | |
| // another class in drug-classes.json and rerun to see a different grouping. | |
| // | |
| // node by-class.mjs prints the markdown section that ledger.md carries | |
| import { readFileSync } from 'node:fs'; | |
| const OUTPUTS = ['Product B: transcript', 'Product B: SOAP', 'Perch: configuration A', 'Perch: configuration B']; | |
| const classes = JSON.parse(readFileSync(new URL('./drug-classes.json', import.meta.url), 'utf8')); | |
| delete classes._comment; | |
| const classOf = new Map(); | |
| for (const [name, drugs] of Object.entries(classes)) for (const d of drugs) classOf.set(d.toLowerCase(), name); | |
| // a cell reads like "found: Baytril; missed: enrofloxacin + false offer: Roflexacin" | |
| const foundIn = (cell) => { | |
| const m = cell.match(/found:\s*([^;+]*)/i); | |
| return new Set(m ? m[1].split(',').map((s) => s.trim().toLowerCase()).filter(Boolean) : []); | |
| }; | |
| const rows = readFileSync(new URL('./ledger.md', import.meta.url), 'utf8').split('\n') | |
| .filter((l) => /^\| [a-z]\d{3} \|/.test(l)) | |
| .map((l) => l.split('|').slice(1, -1).map((s) => s.trim())); | |
| const tally = {}; | |
| for (const name of Object.keys(classes)) tally[name] = { n: 0, out: OUTPUTS.map(() => ({ found: 0, missed: [] })) }; | |
| for (const [id, , truth, ...cells] of rows) { | |
| const drugs = truth.split(',').map((s) => s.trim()).filter((s) => s && !/^\(?none/i.test(s) && s !== '-'); | |
| for (const drug of drugs) { | |
| const cls = classOf.get(drug.toLowerCase()); | |
| if (!cls) throw new Error(`${id}: "${drug}" has no class in drug-classes.json`); | |
| tally[cls].n++; | |
| cells.slice(0, OUTPUTS.length).forEach((cell, i) => { | |
| if (foundIn(cell).has(drug.toLowerCase())) tally[cls].out[i].found++; | |
| else tally[cls].out[i].missed.push(`${drug} (${id})`); | |
| }); | |
| } | |
| } | |
| const lines = []; | |
| lines.push('| drug class | dictated drug instances | ' + OUTPUTS.join(' | ') + ' |'); | |
| lines.push('|---|---|' + OUTPUTS.map(() => '---').join('|') + '|'); | |
| const order = Object.entries(tally).sort((a, b) => b[1].n - a[1].n); | |
| for (const [name, t] of order) lines.push(`| ${name} | ${t.n} | ` + t.out.map((o) => `${o.found} found, ${o.missed.length} missed`).join(' | ') + ' |'); | |
| lines.push(''); | |
| lines.push('Missed drugs, by class and output (clip id in brackets):'); | |
| lines.push(''); | |
| for (const [name, t] of order) { | |
| const parts = t.out.map((o, i) => (o.missed.length ? `${OUTPUTS[i]}: ${o.missed.join(', ')}` : null)).filter(Boolean); | |
| lines.push(`- **${name}.** ${parts.length ? parts.join('. ') + '.' : 'Nothing missed by any output.'}`); | |
| } | |
| console.log(lines.join('\n')); | |