Alexander Bocken
738875e89f
All checks were successful
CI / update (push) Successful in 1m18s
Real-time shopping list with SSE sync between multiple clients, automatic item categorization using embedding-based classification + Bring icon matching, and card-based UI with category grouping. - SSE broadcast for live sync (add/check/remove items across tabs) - Hybrid categorizer: direct catalog lookup → category-scoped embedding search → per-category default icons, with DB caching - 388 Bring catalog icons matched via multilingual-e5-base embeddings - 170+ English→German icon aliases for reliable cross-language matching - Move cospend dashboard to /cospend/dash, /cospend redirects to list - Shopping icon on homepage links to /cospend/list
75 lines
2.5 KiB
TypeScript
75 lines
2.5 KiB
TypeScript
/**
|
|
* Pre-assign each Bring catalog icon to a shopping category using embeddings.
|
|
* This enables category-scoped icon search at runtime.
|
|
*
|
|
* Run: pnpm exec vite-node scripts/assign-icon-categories.ts
|
|
*/
|
|
import { pipeline } from '@huggingface/transformers';
|
|
import { readFileSync, writeFileSync } from 'fs';
|
|
import { resolve } from 'path';
|
|
|
|
const MODEL_NAME = 'Xenova/multilingual-e5-base';
|
|
const CATEGORY_EMBEDDINGS_PATH = resolve('src/lib/data/shoppingCategoryEmbeddings.json');
|
|
const CATALOG_PATH = resolve('static/shopping-icons/catalog.json');
|
|
const OUTPUT_PATH = resolve('src/lib/data/shoppingIconCategories.json');
|
|
|
|
function cosineSimilarity(a: number[], b: number[]): number {
|
|
let dot = 0, normA = 0, normB = 0;
|
|
for (let i = 0; i < a.length; i++) {
|
|
dot += a[i] * b[i];
|
|
normA += a[i] * a[i];
|
|
normB += b[i] * b[i];
|
|
}
|
|
return dot / (Math.sqrt(normA) * Math.sqrt(normB));
|
|
}
|
|
|
|
async function main() {
|
|
const catData = JSON.parse(readFileSync(CATEGORY_EMBEDDINGS_PATH, 'utf-8'));
|
|
const catalog: Record<string, string> = JSON.parse(readFileSync(CATALOG_PATH, 'utf-8'));
|
|
|
|
console.log(`Loading model ${MODEL_NAME}...`);
|
|
const embedder = await pipeline('feature-extraction', MODEL_NAME, { dtype: 'q8' });
|
|
|
|
const iconNames = Object.keys(catalog);
|
|
console.log(`Assigning ${iconNames.length} icons to categories...`);
|
|
|
|
const assignments: Record<string, string> = {};
|
|
|
|
for (let i = 0; i < iconNames.length; i++) {
|
|
const name = iconNames[i];
|
|
const result = await embedder(`query: ${name.toLowerCase()}`, { pooling: 'mean', normalize: true });
|
|
const qv = Array.from(result.data as Float32Array);
|
|
|
|
let bestCategory = 'Sonstiges';
|
|
let bestScore = -1;
|
|
for (const entry of catData.entries) {
|
|
const score = cosineSimilarity(qv, entry.vector);
|
|
if (score > bestScore) {
|
|
bestScore = score;
|
|
bestCategory = entry.category;
|
|
}
|
|
}
|
|
|
|
assignments[name] = bestCategory;
|
|
|
|
if ((i + 1) % 50 === 0) {
|
|
console.log(` ${i + 1}/${iconNames.length}`);
|
|
}
|
|
}
|
|
|
|
writeFileSync(OUTPUT_PATH, JSON.stringify(assignments, null, 2), 'utf-8');
|
|
console.log(`Written ${OUTPUT_PATH} (${iconNames.length} entries)`);
|
|
|
|
// Print summary
|
|
const counts: Record<string, number> = {};
|
|
for (const cat of Object.values(assignments)) {
|
|
counts[cat] = (counts[cat] || 0) + 1;
|
|
}
|
|
console.log('\nCategory distribution:');
|
|
for (const [cat, count] of Object.entries(counts).sort((a, b) => b[1] - a[1])) {
|
|
console.log(` ${cat}: ${count}`);
|
|
}
|
|
}
|
|
|
|
main().catch(console.error);
|