文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 自然科学 > 数学 > BayesSDT Software for Bayesian inference with signal detection theory
侵权投诉

BayesSDT Software for Bayesian inference with signal detection theory

BAYESIAN SDT 453

Table 2

WinBUGS Script for Bayesian SDT Graphical Model

# BAYESIAN SIGNAL DETECTION THEORY#

# INPUT VARIABLES

# H, F, M, C are the hit, false alarm,

# miss and correct rejection signal detection counts#

# OUTPUT VARIABLES

# d, k are discriminability and criterion# h, f are the hit and false-alarm rates# c, b are measures of biasmodel {

# Relating observed counts to underlying Hit and False Alarm rates # Number of Signal Trials is sum of Hit and Miss counts S <- H+M

# Number of Noise Trials is sum of False Alarm and Correct Rejection counts N <- F+C

# Hit counts are Binomial H ~ dbin(h,S)

# False alarm counts are Binomial F ~ dbin(f,N)

# Reparameterization, converting Hit and False Alarm rates # to Discriminability and Bias indices h <- phi(d/2-c) f <- phi(-d/2-c) k <- d/2+c b <- d*c

# These priors over Discriminability and Bias correspond # to uniform priors over the Hit and False Alarm rates MEAND <- 0 MEANC <- 0

LAMBDAD <- 1/2 LAMBDAC <- 2

c ~ dnorm(MEANC,LAMBDAC) d ~ FSDT)}

specified. Additional samples and bins lead to smoother approximations to the posterior density, but come at the expense of additional computing time. As a guide to speed, in the present article, collecting and analyzing 106 samples and 100 bins for the three data sets—many, few, and perfect—took 273 sec on a PC with a processor speed of 2.4 GHz and 2 Gb of RAM.

Figure 4 shows the graphical output of the BayesSDT software, which is the main result of the analysis, although BayesSDT GUI also automatically generates a text file—BayesSDToutput.txt—listing the means and standard devi-ations for each of the parameters. In Figure 4, the posterior distributions for the many data set show that the parameters are estimated with relatively little uncertainty. For the few data set, however, the posterior distributions now represent the much greater degree of uncertainty. For each param-eter, the mode is the same, but the range of possible values the parameter could take is greater. This is a natural con-sequence of the parameter estimates being based on fewer data, and it is well handled by the Bayesian approach. For the perfect data set, the modal hit and false alarm rates are 1.0 and 0.0, but other possibilities have some density; thus, the remaining discriminability, bias, and criterion pa-

rameters are all well defined without the need for edge corrections. It is also worth noting that these hit and false alarm posteriors do not take the Gaussian form assumed by standard analyses, but instead are sensitive to the theoreti-cal bounds on the values of these rate parameters.

MATLAB function. Rather than using the BayesSDT GUI, it is possible to call the BayesSDT function directly. This is especially useful for analyzing more than four data sets simultaneously. The BayesSDT function takes a single argument, D, which is a structured variable with 14 fields.D.ndatasets is the number of data sets to be analyzed.

D.sdt. This is a matrix with the rows represent-ing data sets and the columns representing signal detection data counts. The four columns correspond to hits, false alarms, misses, and correct rejections, respectively.

D.nsamples is the number of posterior samples to generate.

D.nbins is the number of bins to use in drawing histograms of the posterior densities.

上一页第4页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email