MokuMoku Town 🏙️
A virtual office in your browser. Loosely connected, deeply focused.

LightGBM-MoE 🌲
A regime-switching / Mixture-of-Experts extension of LightGBM.

• #2321 — perf(distributions): use vmap for the binomial sampler on CPU as well (under review)
• #2320 — perf(util): avoid double allocation of the fori_collect collection
• #2319 — perf(infer): avoid keeping two copies of single-chain MCMC samples (under review)
• #2318 — Skip the zero-width pad in soft_vmap
• #2317 — perf(util): avoid per-step host sync from the diagnostics string in the MCMC progress-bar loop (under review)
• #2251 — Speed up posterior diagnostics (summary 1.7x, effective_sample_size 2.1x)
• #2250 — Drop the masked logsumexp reduction in mixture log_prob
• #2248 — Fix generator being appended instead of a list in scan carry_shapes
• #2247 — Avoid per-step host sync in the SVI progress-bar loop
• #2222 — Removed unnecessary deep copies in module sampling and batch-shape promotion

• #7483 — [c++] avoid copying per-row contribution maps in sparse prediction output (under review)
• #7482 — [python-package] keep float32 feature matrix when pandas categorical columns contain missing values (under review)
• #7247 — Added LightGBM-MoE to the official external repositories list
• #7246 — Added native int8 input support for pre-discretized features (under review)