sweep.hpp File Reference

Parameter sweep (8.4/8.5): cartesian grid over strategy parameters run on a std::jthread pool; every worker owns its own data cursor (source_factory) and backend, results come back in grid order (first parameter varies slowest). More...

#include "fastmm/backtest/backtest_runner.hpp"
#include <cstddef>
#include <cstdint>
#include <functional>
#include <limits>
#include <memory>
#include <string>
#include <string_view>
#include <utility>
#include <vector>

Go to the source code of this file.

Classes
struct	fastmm::bt::SweepPoint
struct	fastmm::bt::WalkForwardFold
struct	fastmm::bt::WalkForwardReport

Namespaces
namespace	fastmm
namespace	fastmm::bt

Typedefs
using	fastmm::bt::ParamGrid = std::vector< std::pair< std::string, std::vector< std::string > > >
using	fastmm::bt::SourceFactory = std::function< std::unique_ptr< MdSource >()>
	may return nullptr
using	fastmm::bt::RunFn = std::function< BacktestResult(const BacktestConfig &, MdSource *)>
using	fastmm::bt::ScoreFn = double(*)(const BacktestResult &)
	Ranking metric of a run, higher is better: "net_pnl", "realized_pnl", "sharpe_bar" or "spread_captured_bps".

Functions
std::vector< ParamMap >	fastmm::bt::expand_grid (const ParamGrid &grid)
	Expands the grid in order: [{k1: v11, k2: v21}, {k1: v11, k2: v22}, ...].
std::vector< SweepPoint >	fastmm::bt::sweep (const BacktestConfig &base, const ParamGrid &grid, const SourceFactory &source_factory, const RunFn &run, int threads=0)
	Generic driver; run is called once per point on a worker thread.
template<StrategyLike S>
std::vector< SweepPoint >	fastmm::bt::sweep (const BacktestConfig &base, const ParamGrid &grid, const SourceFactory &source_factory={}, int threads=0)
std::vector< SweepPoint >	fastmm::bt::sweep_by_name (const BacktestConfig &base, std::string strategy, const ParamGrid &grid, const SourceFactory &source_factory={}, int threads=0)
ScoreFn	fastmm::bt::score_metric (std::string_view name)
WalkForwardReport	fastmm::bt::walk_forward (const BacktestConfig &base, const ParamGrid &grid, const SourceFactory &source_factory, const RunFn &run, int folds, std::string_view metric="net_pnl", int threads=0)
	Splits the data's event-time range into folds equal slices and runs the whole grid on each one (a fresh backtest per point and fold, all on one pool of threads workers).
WalkForwardReport	fastmm::bt::walk_forward_by_name (const BacktestConfig &base, std::string strategy, const ParamGrid &grid, const SourceFactory &source_factory, int folds, std::string_view metric="net_pnl", int threads=0)

Detailed Description

Parameter sweep (8.4/8.5): cartesian grid over strategy parameters run on a std::jthread pool; every worker owns its own data cursor (source_factory) and backend, results come back in grid order (first parameter varies slowest).

walk_forward() runs the same grid on K consecutive time slices (folds) of the data and scores the in-sample winner of fold i-1 on fold i: an out-of-sample check of the sweep's best row.