Fixed throughout. Where a symbol appears in the code under a different name,
the code's name is given.
Symbol Is In the code
x x x A patch term — a tree in the typed grammar PatchTreep p p A tree path , e.g. node/0/1 path keys t t t A trace — the execution record of the grammar program fugue::Traceφ ( x ) ∈ R 40 \varphi(x) \in \R^{40} φ ( x ) ∈ R 40 The feature vector of x x x Features::phi()φ audio ∈ R 15 \varphi_{\text{audio}} \in \R^{15} φ audio ∈ R 15 Perceptual descriptors of the render AudioFeaturesφ struct ∈ R 25 \varphi_{\text{struct}} \in \R^{25} φ struct ∈ R 25 Structural descriptors of the term StructFeaturesz z z A standardized feature vector, z = ( φ − μ ) / s z = (\varphi - \mu)/s z = ( φ − μ ) / s phi_std
φ \varphi φ [ φ audio ; φ struct ] [\varphi_{\text{audio}} ; \varphi_{\text{struct}}] [ φ audio ; φ struct ] φ \varphi φ ϕ \phi ϕ phi is this vector.
Symbol Is In the code
K K K Number of style lenses (≤ 5 \le 5 ≤ 5 TasteConfig::k_stylesd d d Feature dimension (40 40 40 TasteConfig::n_featuresθ k ∈ R d \theta_k \in \R^{d} θ k ∈ R d Lens k k k TasteSample::theta[k]θ \theta θ All of them, K × d K \times d K × d TasteSample::thetau ( x ) u(x) u ( x ) Latent utility of x x x utility_mixu k ( x ) u_k(x) u k ( x ) Lens k k k θ k ⊤ z \theta_k^\top z θ k ⊤ z utility(phi, k)τ s \tau_s τ s Session s s s TasteSample::tau[s]c j c_j c j Star cutpoint j j j TasteSample::cuts[j]σ θ \sigma_\theta σ θ Prior SD of one θ \theta θ TasteConfig::sigma_theta()s K s_K s K Max-of-K K K MAX_NORMAL_SDS S S Number of sessions in the log FitSet::n_sessions()
Symbol Is In the code
p grammar ( x ) p_{\text{grammar}}(x) p grammar ( x ) Prior probability of term x x x PatchGrammarPriorβ \beta β Boltzmann sharpness SessionConfig::betaπ β \pi_\beta π β The target, ∝ p grammar ( x ) e β E [ u θ ( x ) ] \propto p_{\text{grammar}}(x),e^{\beta,\E[u_\theta(x)]} ∝ p grammar ( x ) e β E [ u θ ( x )] — η \eta η Proposal-tilt strength SessionConfig::proposal_tiltL \mathcal{L} L The set of locked addresses locked: HashSet<String>
σ ( ⋅ ) \sigma(\cdot) σ ( ⋅ ) logistic function σ ( v ) = 1 / ( 1 + e − v ) \sigma(v) = 1/(1+e^{-v}) σ ( v ) = 1/ ( 1 + e − v ) σ θ \sigma_\theta σ θ s s s log \log log nats .Indices are 0-based , matching the code, including cutpoint indices, which
matters for reading the ordinal
likelihood .
Weights w w w normalized unless stated: importance weights sum
to one, recency weights are relative to the newest observation being 1 1 1
"Standardized" always means after the affine transform in
Standardization . The taste model never sees
raw φ \varphi φ log never stores anything else.
Defined in www/reference/book.toml so a symbol cannot mean two things on two
pages:
Macro Renders
\RR \R R \EE \E E \phivecφ \phivec φ \thetakθ k \thetak θ k \sigσ \sig σ