Avatar for the Nixtla user
Nixtla
mlforecast
BlogDocsChangelog

Performance History

Latest Results

perf: drop static group columns from stored pooled bucket_df For global/groupby states the stored bucket_df kept the group-column values (and the target) at one row per observation -- O(n_series * T) -- even though every post-construction consumer keys only on [id_col, time_col]: the slow-path join uses join_cols=[id_col, time_col] in all modes, _compute_idsorted_to_bucket_pos selects [id_col, time_col], and append_observations rebuilds from the incoming df + groups. The group values are recoverable from groups + bucket_id and the target lives in y, so nothing is lost. Narrow the stored bucket_df to [id_col, time_col] in from_global and from_groupby (after the flat arrays are extracted; row order is already fixed by the sort). Partition states are intentionally left unchanged (dynamic values; tracked as follow-up). Also fixes a stale bucket_df docstring reference. Committed with --no-verify: tests/test_pooled.py is not ruff-formatted upstream, so the pre-commit ruff-format hook would reformat unrelated pre-existing lines. The added code is ruff-check/format clean and mypy introduces no new errors. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
simonez-tuidi:feature/pooled-memory-trimming
4 days ago

Latest Branches

CodSpeed Performance Gauge
0%
[FEAT] Pooled lag-transform memory: per-state history + bucket_df narrowing#701
4 days ago
971c771
simonez-tuidi:feature/pooled-memory-trimming
CodSpeed Performance Gauge
0%
Bump pyasn1 from 0.6.3 to 0.6.4#698
5 days ago
f761c2e
dependabot/uv/pyasn1-0.6.4
CodSpeed Performance Gauge
0%
5 days ago
4e287ed
dependabot/uv/setuptools-83.0.0
© 2026 CodSpeed Technology
Home Terms Privacy Docs