Index of /~carterja/code

      Name                    Last modified       Size  Description


[DIR] Parent Directory        20-Sep-2009 10:46      -  
[   ] filterredwv.pro         02-Sep-2009 16:25     1k  
[   ] minfuncwv.pro           02-Sep-2009 16:25     1k  
[   ] solveredwv.pro          03-Sep-2009 11:17     5k  
[   ] waveletlike.pro         02-Sep-2009 16:25     1k  

WAVELET-BASED METHOD FOR PARAMETER ESTIMATION; IDL CODE

- For details refer to the paper by Carter & Winn (2009):

	Parameter Estimation from Time-Series Data with Correlated Errors:
	A Wavelet-Based Method an its Application to Transit Light Curves


- Quick overview of IDL procedures or functions (refer to actual code for implementation
	details, calling sequence, examples, etc.):

waveletlike - 	computes the "wavelet (log) likelihood" for a data vector
		with 2^J elements for J a positive integer.  The 1/f+white noise
		model is parameterized by two parameters (SIGMA_R, SIGMA_W) that are
		related to the standard deviation of the two components.
filterredwv -	Filters out the 1/f component for a data vector with 2^J elements 
		for model parameters SIGMA_R and SIGMA_W.
solveredwv  -	Uses AMOEBA to solve for model parameters SIGMA_R and SIGMA_W given
		a data vector with 2^J elements.  Also separates 1/f and white components
		into two separate output vectors.
minfuncwv   -	Wrapper function for use with solveredwv.  Is not a stand-alone function.

- Adding models for parameter estimation

	Our code here only solves for the noise parameters SIGMA_R and SIGMA_W,
	however, the tools necessary for parameter estimation with additional models
	have been provided.

	The function waveletlike can be used to assess the likelihood of a set of residuals
	after having removed a guess for any deterministic signal in the data.  For
	example, a model transit light curve can first be subtracted from light curve data
	before calculating log(likelihood) with waveletlike.  One may then use standard
	estimation techniques to determine noise and transit parameters; e.g., via AMOEBA 
	minimization of -log(likelihood) or Markov chain Monte Carlo where the likelihood,
	calculated as a log in waveletlike, is used as the jump-transition likelihood.

	Similarly, filterredwv can be used to "whiten" residuals after having removed 
	the best fit deterministic signal.

- Revisions
	Written,	Josh Carter 	September 2009