fastmm::research::FeatureConfig Struct Reference

#include <fastmm/research/extractor.hpp>

Public Attributes
std::vector< Duration >	horizons
	Forward horizons.
std::size_t	imbalance_levels = 1
	Levels summed by fastmm::imbalance(). 1 is the touch, which is all a top-of-book feed shows.
Duration	subsample {}
	Minimum spacing between rows of one instrument.
bool	sample_book_updates = true
	What emits a row.
bool	sample_trades = false
std::uint32_t	instruments = 1
	Books kept. Events for a higher instrument id are counted and ignored.

Member Data Documentation

horizons

std::vector<Duration> fastmm::research::FeatureConfig::horizons

Forward horizons.

Empty selects default_horizons(); non-positive entries are dropped and the rest are sorted ascending and de-duplicated.

imbalance_levels

std::size_t fastmm::research::FeatureConfig::imbalance_levels = 1

Levels summed by fastmm::imbalance(). 1 is the touch, which is all a top-of-book feed shows.

subsample

Duration fastmm::research::FeatureConfig::subsample {}

Minimum spacing between rows of one instrument.

Zero samples every book update; 10 ms keeps at most one row per instrument per 10 ms of event time.

sample_book_updates

bool fastmm::research::FeatureConfig::sample_book_updates = true

What emits a row.

A book update emits one after it has been applied; a trade emits one with the book as it stood when the trade printed. Trades off by default; with book updates off and trades on, the table has exactly one row per trade, at the timestamp of every fill a backtest over the same data can produce, which is what the cross-check against fills.csv needs. Turning both off leaves an empty table.

sample_trades

bool fastmm::research::FeatureConfig::sample_trades = false

instruments

std::uint32_t fastmm::research::FeatureConfig::instruments = 1

Books kept. Events for a higher instrument id are counted and ignored.