I put a 10,000-species bird sound model into my React Native app. Wrote up how.
The license trap, 409 MB to 105 MB, and why raw scores aren't probabilities. https://t.co/QeBMdKwiRj
Crossed 100 followers! Currently at 168 🚀
+130 new connections in just 2 days. The spike was wild.
Great meeting so many builders here. Think we can hit 200 today? 👀
Hey @X algorithm 🤖
I'm looking to #CONNECT with people who are:
React Native developers 📱
Solo founders & indie hackers
Building cool mobile apps
Tech enthusiasts
Let’s connect! Drop a line or what you’re building below 👇
Shipping an ML model in a paid app? Check the license of the weights, not just the repo.
The best-known bird sound model, from Cornell: code MIT, weights CC BY-NC-SA. No commercial use.
I went with an open model from Google instead. Apache 2.0, ~10,000 bird species.
@098uyiubgvgcss Good catch, that would bite with raw logits. I softmax over the bird classes first, then multiply those probabilities by the country weight, so it's all positive. Skipped that step in the article.
I put a 10,000-species bird sound model into my React Native app. Wrote up how.
The license trap, 409 MB to 105 MB, and why raw scores aren't probabilities. https://t.co/QeBMdKwiRj