Date: Thu, 28 Mar 2024 23:30:48 +0100 (CET) Message-ID: <1411607857.2809.1711665048772@localhost> Subject: Exported From Confluence MIME-Version: 1.0 Content-Type: multipart/related; boundary="----=_Part_2808_223516339.1711665048770" ------=_Part_2808_223516339.1711665048770 Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Content-Location: file:///C:/exported.html
Parameter Name |
Default Value |
Parameter Range |
Description |
---|---|---|---|
FILTERING |
false | {true, false} | switches filtering for fragment sizes on/off |
SIZE_DISTRIBUTION |
specifies the probability distribution over f= ragment sizes to be retained in the final set. The first possibility to pro= vide a distribution is by a string that describes parameters of a normal di= stribution in the form N(mean, sd),e.g., N(800, 200); the alternative is to= provide as parameter value the name of a file that contains an empirical d= istribution. |
||
SIZE_SAMPLING |
AC | {RJ,AC,MH} | identifies the method used for sub-sampling f= ragments: either rejection-sampling (RJ), a variant of rejection-sampling w= ith a minimal rejection rate (so-called acceptance sampling, AC), or the we= ll known Metropolis-Hastings algorithm (MH) can be selected |
If a size filtering step is carried out (parameter FILTERING), each frag= ment gets first assigned a probability according to the provided distributi= on (SIZE_DISTRIBUTION); either normal distributions can be characterized by= their characteristic attributes "mean" and "standard deviation", or empiri= c distribution can be provided in the form of a file. Subsequently, fragmen= ts are selected according to one of the following sub-sampling algorithms:<= /p>
Rejection Sampling (RJ) - a Bernoulli trial is carried out dire= ctly against the probability assigned by the provided distribution, which t= hen decides whether the fragment is retained or discarded. Fragment sizes o= btained by this algorithm distribute as the distribution specified by SIZE_= DISTRIBUTION.
Acceptance Sampling (AC) - a modification of rejection sampling= which stretches the maximum value of the probabilities in SIZE_DISTRIBUTIO= N to 1 before applying rejection filtering. The method shows a higher yield= of retained fragments, and largely preserves the characteristics of the pr= ovided size distribution.
Metropolis Hastings (MH) - a Montecarlo Markov Chain algorithm = is employed in the sub-sampling process. Iteratively applied assumptions on= the posterior distribution may distort the shape of the distribution, espe= cially if the initial fragment size distribution and the provided filter di= stribution (SIZE_DISTRIBUTION) differ significantly from each other. Howeve= r--at these costs--the algorithm looses less fragments than RJ or AC.