pub struct StyleShareRecord {
pub observations: usize,
pub k: usize,
pub shares: Vec<f64>,
}Expand description
What one posterior fit claimed for each style lens.
Recorded per fit and persisted, because the question it answers is about
real sessions over time and cannot be answered from one of them.
SessionConfig::k_styles documents an option that is deliberately not
taken — cap K at 3, taking the fit from 206 sites to 126 for a ~1.6×
mature-fit win — and gates it explicitly on whether lenses 4 and 5 sit near
zero share across real sessions. Nothing collected that, so the decision
could not be made either way; this is the collection.
It is deliberately not a judgment. A row says what the shares were at a
given evidence count, and how many lenses the fit was even allowed (k
grows with the log and is capped by config, so an early row with two lenses
is not evidence that lenses 3–5 are idle — it is evidence they did not
exist yet). Reading rows where k == k_styles is what the option needs.
Fields§
§observations: usizeObservations in the log at the time of the fit.
k: usizeLenses this fit was allowed — min(1 + log/20, k_styles).
Share of the pool claimed by each lens, aligned, summing to ~1.
Trait Implementations§
Source§fn clone(&self) -> StyleShareRecord
fn clone(&self) -> StyleShareRecord
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more