Package com.quantfinlib.persist
Checkpoint writes one binary file of
named sections at end of day and restores it at session start. The write
goes to a temporary file that is forced to disk and renamed, so a crash
mid-save leaves the previous file intact.
The models that learn across days expose
writeState(DataOutput) / readState(DataInput) pairs:
the seasonality trio (microstructure.VolumeCurve,
VolatilityCurve, SpreadForecaster),
microstructure.OnlineAlphaLearner (weights and the out-of-sample
IC evidence),
microstructure.LeadLagEstimator,
microstructure.EwmaCovariance (the basket risk matrix),
microstructure.KylesLambda (learned depth),
microstructure.ClosingAuctionModel (learned auction share),
microstructure.AlphaEnsemble (per-component IC evidence),
rfq.RfqDealerScorecard (learned dealer-panel quality), and the
venue-quality cards (execution.VenueScorecard, format v2 with
fill markouts, still reads v1; and fx.LpScorecard).
Intraday state resets on read; configuration (bucket/venue counts) must
match or the read throws. HiddenLiquidityDetector is
not persistable: its state is keyed by price level, and overnight the
price ladder moves. Restoring it would attribute yesterday's icebergs
to today's unrelated levels.
4.0. This package stays in the stable jar in 4.0. The 3.1 design
proposed moving it out, but stable packages (crb, fx, execution,
microstructure) depend on it, and a move would make the module graph
cyclic. See docs/DESIGN_4.0.md.
-
ClassDescriptionMulti-day persistence of learned state.Random access to a loaded checkpoint's sections by name.A model's state deserializer — typically a
readStatereference.A model's state serializer — typically awriteStatereference.Collects named sections and commits them atomically on close.