Content-type: text/html Man page of X-GEN

X-GEN

Section: X-GEN Commands (1)
Updated: April 2005
Index Return to Main Contents
 

NAME

X-GEN - reject  

DESCRIPTION

"reject" flags certain observations found in a merged set of data as being outliers. The outliers are noted on the basis of disagreement with the mean for the reflection involved according to
Q = |Iij - gij * <Ij >| / sij ,

where Iij is the intensity of the ith observation of the jth reflection, gij is the scaling-function value associated with it, <Ij> is the mean intensity for all observations of reflection j, and sij is the standard deviation associated with Iij . Thus if Q is greater than a certain cutoff, an observation will be flagged for deletion. The cutoff value can be determined several ways, as described below.

-a
Set up "Curses" menus for looking at difficult cases. This used to be the default, and no longer is.
-b
This specifies a "batch mode", in which all decisions about eliminating outliers is performed by the software rather than leaving any of the decisions up to the user. This is now the default, but didn't used to be.
-c
This specifies creation of a detailed log file with environment variable name CHECKREFL into which a reasonably-formatted list of all the observations of any suspect reflection are written, together with an accounting of how the program handles that reflection.
-g
For flagged reflections with exactly two observations, keep the observation that has the smaller sigma value.
-i
For flagged reflections with exactly two observations, keep both observations.
-k
For flagged reflections with exactly two observations, delete both observations.
-l
For flagged reflections with exactly two observations, keep the observation with the larger intensity.
-m
Inflate the sigmas of all observations by a machine-error term. This machine-error term is applied only during the outlier-rejection effort. The extra component of the variance is set to (0.025 * I) squared.
-n
Separate Bijvoet-related reflections for purposes of computing mean intensities and calculating deviations from those means. Thus with this flag on the (3,6,4) reflection will be treated as independent of the (-3,-6,-4).
-o
Merge Bijvoet-related reflections for purposes of computing mean intensities and calculating deviations from those means. Thus with this flag on the (3,6,4) reflection will be treated as equivalent to the (-3,-6,-4). This is the default behavior of the program.
-p
If this Boolean is on, observations that had been flagged for deletion in a previous run of "reject" will remain flagged this time around. Otherwise these observations will be restored into consideration.
-s
For flagged reflections with exactly two observations, when the user is running in Curses (-a) mode, the program will display both observations and let the user decide how to handle the reflection.
-u
Do not inflate the observation sigmas with a machine error. This is the converse of the -m flag.
-v
If this Boolean is on, a moderate amount of additional diagnostics beyond the default will be written to the XLOG log file and stdout.
-f<val>
This specifies the approximate fraction of the data that
are to be flagged for deletion. Thus if 0.03 is specified here, then roughly 3% of the observations will be flagged for deletion. The fraction is always approximate because it is used to establish the cutoff |I-<I>|/sigma criterion, and the actual number deleted is obtained from a different algorithm based on that cutoff. Default: this technique is only exercised if "-f" is specified with a legal value of <val> (i.e. a value between 0 and 1).
-q<val>
This value specifies the cutoff deviation in (I - <I>) /
sigma above which an observation will be considered for deletion. If no value is specified the program computes a cutoff as a slowly-increasing function of the number of observations in the dataset; if the dataset contains 10000 observations the default cutoff is 3.92; with 200000 it is 4.58.
-r < h k l >
If the program winds up flagging a very large
number of observations for deletion, the buffer into which those observations are put may fill up. Under those circumstances the program will exit after it has handled as many observations as it can and will report the last (h,k,l) value that it dealt with. A subsequent run of reject in which the user specifies that (h,k,l) value will pick up the remaining data:
reject -q5.4 -b -r 8 11 13
will begin the process at the (8,11,13) reflection. This feature has not been extensively tested in recent years and may no longer work.
-t<val>
This value specifies the target Rmerge value (expressed as
a fraction, not a percentage) that the algorithm will attempt to achieve. If the value 0.05 is specified, then the program will set the cutoff |I-<I>|/sigma criterion such that enough deletions are done to achieve an Rmerge of 0.05 (5%). As the car manufacturers would say, your mileage may vary. Default: this technique is only exercised if "-t" is specified with a legal value of <val> (i.e. a value between 0 and 1).
 

REPORTING BUGS

Report bugs to Andy Howard at howard@iit.edu or 312-567-5881.  

COPYRIGHT

Copyright © 2002, Illinois Institute of Technology. See the file 'LICENSE' for information on usage and redistribution of this file, and for a DISCLAIMER OF ALL WARRANTIES


 

Index

NAME
DESCRIPTION
REPORTING BUGS
COPYRIGHT

This document was created by man2html, using the manual pages.
Time: 02:08:09 GMT, October 03, 2005