Aravindh Raman
0e4ca2d35e
addressing scikit-learn deserialization of untrusted data (roots to unsafe joblib.load, https://github.com/brave/brave-core/security/dependabot/92) and reproducibility concern raised by code scanner (https://github.com/brave/brave-core/security/code-scanning/147) the previously generated parameters (found in browser/bandwidth_linreg_parameters.h) will be used for the estimations.
914 B
914 B
Brave Performance Predictor
This component is based on our blog post.
The current implementation is only for bandwidth prediction and uses a subset of the features discussed in the post for better cross-platform support. Key groups of features used are:
- URLs of blocked requests, from which we extract counts and distinct third-parties (e.g. tracking networks)
- total counts and sizes of loaded requests by type (scripts, images, fonts, stylesheets, media, document and “other”)
- basic page performance metrics (page load time, page interactive time, dom content loaded, first visual change, first meaningful paint)
The component includes:
- Generated linear regression model to parameters in a C++ header file (bandwidth_linreg_parameters.h)
- An interface to the model that buffers submitted features and runs the model when requested