Geen omschrijving

jherve d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
src d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
static cc5966b339 More tweaks to UI 1 jaar geleden
templates 67e0e31efd Slight tweaks to UI 1 jaar geleden
tests 1c91c4bfe6 Initial commit 1 jaar geleden
.gitignore d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
README.md 7e076bd9d7 Add a note about a bug 1 jaar geleden
config.py caa854cba8 Format some file 1 jaar geleden
pyproject.toml d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
requirements-dev.lock d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
requirements-embeddings.lock 5d907e27a8 Add a requirements file for embeddings 1 jaar geleden
requirements.lock d51ff03bb2 Switch to Annoy for vector search/indexing 1 jaar geleden
settings.toml 207cc110db Move database URL to secrets 1 jaar geleden

README.md

de_quoi_parle_le_monde

Bug

In the featured_article_snapshot_id view, the field featured_article_snapshot_id is taken as if it was unique by row, but it is not.

This can be easily checked with this query :

SELECT * FROM (
    SELECT featured_article_snapshot_id, json_group_array(snapshot_id), COUNT(*) as count
    FROM snapshot_apparitions
    WHERE is_main -- Not required
    GROUP BY featured_article_snapshot_id
)
WHERE count > 1

Among other things it leads to "deadends" while browsing the UI, likely because the timestamp search and time diff relies on this false assumption.