⚡ Bolt: vectorize BasicEstimator.predict#32
Conversation
Optimized the `predict` method of `BasicEstimator` by vectorizing the distance calculation using the squared distance expansion formula. - Replaced the Python loop over input embeddings with a vectorized matrix operation using `np.dot` and broadcasting. - Pre-calculated squared norms of fitted embeddings in `fit` to further speed up predictions. - Maintained backward compatibility for older saved models without pre-calculated norms using `getattr` fallback. - Guarded against floating-point precision errors using `np.maximum(..., 0)`. - Achieved a ~7.5x speedup for batches of 500 queries against 2000 fitted embeddings. Co-authored-by: guesswh0 <10531675+guesswh0@users.noreply.github.com>
|
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request. When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down. I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job! For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with New to Jules? Learn more at jules.google/docs. For security, I will only act on instructions from the user who triggered this task. |
1 participant
⚡ Bolt: vectorize BasicEstimator.predict
💡 What: Optimized the
predictmethod ofBasicEstimatorinface_engine/models/basic_estimator.py.🎯 Why: The previous implementation used a Python loop over input embeddings, calculating distances one by one, which was inefficient for batch predictions.
📊 Impact: Reduces batch prediction time by ~85% (from ~0.26s to ~0.035s for 500x2000 comparison, a ~7.5x speedup).
🔬 Measurement: Verified with a custom benchmark script (now removed) comparing old loop-based logic against the new vectorized matrix operation, and confirmed numerical equivalence within
rtol=1e-4.PR created automatically by Jules for task 4389138148714191695 started by @guesswh0