ABC Class Reference

Model estimation by Approximate Bayesian Computation. More...

#include <ABC.hpp>

Public Types
enum	TransformMode { NONE, LOG, TAN }
	Modes for parameter transformation.

Public Member Functions
	ABC ()
	Constructor.
	~ABC ()
	Destructor.
void	number_of_statistics (unsigned int ns)
	Sets number of statistics. More...
void	add_fname (const char *fname, unsigned int number_of_params)
	Adds a file name.
unsigned int	number_of_samples () const
	Gets number of imported data samples.
void	get_threshold (double tolerance)
	Gets regression threshold. More...
double	sd (unsigned int index) const
	Gets a standard deviation. More...
void	obs (unsigned int index, double value)
	Sets a summary statistics. More...
unsigned int	rejection (const char *outfname, bool exportlabels=false)
	Performs rejection step. More...
unsigned int	regression (const char *infname, const char *outfname, TransformMode mode, const char *header="")
	Performs regression step. More...
double	threshold () const
	Gets Euclidean threshold.

Detailed Description

Model estimation by Approximate Bayesian Computation.

It is required to set the number of statistics and at least one input file name before performing analysis. The analysis itself consists in several steps. (1) Computation of the threshold, which requires to read through all files and imports statistics. In the process, the standard deviation of all statistics will be calculated and will be available. (2) Computation of Euclidean distances and weights and generation of a second sample file with weights and (non-standardized) statistics with only non-null weights exported. This step requires that the observed statistics have been set (between steps 1 and 2). (3) Local-linear regression using the fit method. While in the previous steps several models can be mixed, in this step a single model can be processed at a time. The output is a simple file with adjusted parameters only.

Member Function Documentation

void get_threshold

(

double

tolerance

)

Gets regression threshold.

If several files are loaded, the data will be aggregated (note that they must all contain the same number of statistics). At least one file must have been set, and the number of statistics must have been set as well.

Parameters

tolerance

rejection threshold (proportion of points in the local region.

void number_of_statistics

(

unsigned int

ns

)

Sets number of statistics.

If data was already present in the instance, it will all be cleared.

void obs	(	unsigned int	index,
		double	value
	)

Sets a summary statistics.

The number of statistics must have been set, and the index must not be out of bound.

unsigned int regression	(	const char *	infname,
		const char *	outfname,
		TransformMode	mode,
		const char *	header = ""
	)

Performs regression step.

Parameters

infname	input file name (generated using rejection)
outfname	output file name (final posterior)
mode	transformation mode
header	outfile file header (name of parameters; if empty string, no header is printed)

Returns

Number of data point processed

unsigned int rejection	(	const char *	outfname,
		bool	exportlabels = false
	)

Performs rejection step.

The observed value must have been entered (otherwise the results will be meaningless), and the threshold must have been computed.

Parameters

outfname	the name of the intermediary file.
exportlabels	if true: exports a tag at the beginning of each line to identify the file or origin of each accepted sample (starting from 1).

Returns

the number of points in the local region.

double sd

(

unsigned int

index

)

const

Gets a standard deviation.

The threshold() method must have been called, and the index must not be out of bound.

---

The documentation for this class was generated from the following files:

• /home/stephane/Documents/devel/egglib2/egglib-cpp/ABC.hpp
• /home/stephane/Documents/devel/egglib2/egglib-cpp/ABC.cpp