You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
get_history now caches per-DataFrame numpy views of price columns instead of rebuilding a pandas Series for every requested field and stock, and reuses the already-built int64 date index for window alignment. Multi-stock history queries run substantially faster with identical results.
Column aliases (money/price) are normalized once when daily or minute data is loaded, instead of copying the source frame inside get_history on every call. The alias rule now lives in a single storage.ensure_alias_columns helper.
Date columns that are already datetime-typed skip the redundant pd.to_datetime pass during data loading, shortening startup for large datasets.
Fixed
Cached column views are keyed by DataFrame identity and guarded by a weak reference plus the data version, so recycled object ids or replaced data sources can no longer return values from a previous frame.
Tests
Added regression coverage for recycled DataFrame ids and for eviction of cached views when the source frame is collected.