 |
ATLAS Offline Software
|
Loading...
Searching...
No Matches
|
ATLAS Offline Software
|
Variables | |
| parser = argparse.ArgumentParser(usage=__doc__) | |
| nargs | |
| help | |
| default | |
| dest | |
| metavar | |
| action | |
| False | |
| True | |
| args = parser.parse_args() | |
| S1D_MODE | |
| S2D_MODE | |
| S3D_MODE | |
| dict | analysisobjects_in = {} |
| filename | |
| scale = float(scale) | |
| aos = yoda.read(filename, patterns=args.PATHPATTERNS, unpatterns=args.PATHUNPATTERNS) | |
| dict | analysisobjects_out = {} |
| aotype = type(aos[0]) | |
| str | msg = "WARNING: cannot merge mismatched analysis object types for path %s: " % p |
| bool | scatter_fail = False |
| list | saos = [] |
| sao = ao.mkScatter() | |
| saotype = type(saos[0]) | |
| list | aos_nonzero = [ao for ao in aos if ao.sumW() != 0] |
| rescale = None | |
| ao_out = aos[0].clone() | |
| dim = ao_out.dim() | |
| SND_MODE = getattr(args, "S%dD_MODE" % dim) | |
| list | axis = ['', 'x', 'y', 'z'] |
| npoints = len(ao_out.points()) | |
| float | val_i = 0.0 |
| dict | ep_i = {} |
| dict | em_i = {} |
| variations = ao.variations() | |
\
%(prog)s [-o outfile] <yodafile1>[:<scale1>] <yodafile2>[:<scale1>] ...
e.g. %(prog)s run1.yoda run2.yoda run3.yoda (unweighted merging of three runs)
or %(prog)s run1.yoda:2.0 run2.yoda:3.142 (weighted merging of two runs)
Merge analysis objects from multiple YODA files, combining the statistics of
objects whose names are found in multiple files. May be used either to merge
disjoint collections of data objects, or to combine multiple statistically
independent runs of the same data objects into one high-statistics run. Optional
scaling parameters may be given to rescale the weights of the objects on a
per-file basis before merging.
By default the output is written to stdout since we can't guess what would be
a good automatic filename choice! Use the -o option to provide an output filename.
IMPORTANT!
This script is not meant to handle all run merging situations or data objects:
there are limitations to what can be inferred from data objects alone. If you
need to do something more complex than the common cases handled by this script,
please write your own script / program to load and process the data objects.
SCATTERS (E.G. HISTOGRAM RATIOS) CAN'T BE MERGED
Note that 'scatter' data objects, as opposed to histograms, cannot be merged
by this tool since they do not preserve sufficient statistical
information. The canonical example of this is a ratio plot: there are
infinitely many combinations of numerator and denominator which could give the
same ratio, and the result does not indicate anything about which of those
infinite inputs is right (or the effects of correlations in the division).
If you need to merge Scatter2D objects, you can write your own Python script
or C++ program using the YODA interface, and apply whatever case-specific
treatment is appropriate. By default the first such copy encountered will be
returned as the 'merged' output, with no actual merging having been done.
NORMALIZED, UNNORMALIZED, OR A MIX?
An important detail in histogram merging is whether a statistical treatment
for normalized or unnormalized histograms should be used: in the former case
the normalization scaling must be undone *before* the histograms are added
together, and then re-applied afterwards. This script examines the ScaledBy
attribute each histograms to determine if it has been normalized. We make the
assumption that if ScaledBy exists (i.e. h.scaleW has been called) then the
histogram is normalized and we normalize the resulting merged histogram to the
weighted average of input norms; if there is no ScaledBy, we assume that the
histogram is not normalised.
This is not an infallible approach, but we believe is more robust than heuristics
to determine whether norms are sufficiently close to be considered equal.
In complicated situations you will again be better off writing your own
script or program to do the merging: the merging machinery of this script is
available directly in the yoda Python module.
MORE NOTES
If all the input histograms with a particular path are found to have the same
normalization, and they have ScaledBy attributes indicating that a histogram
weight scaling has been applied in producing the input histograms, each
histogram in that group will be first unscaled by their appropriate factor, then
merged, and then re-normalized to the target value. Otherwise the weights from
each histogram copy will be directly added together with no attempt to guess an
appropriate normalization. The normalization guesses (and they are guesses --
see below) are made *before* application of the per-file scaling arguments.
IMPORTANT: note from the above that this script can't work out what to do
re. scaling and normalization of output histograms from the input data files
alone. It may be possible (although unlikely) that input histograms have the
same normalization but are meant to be added directly. It may also be the case
(and much more likely) that histograms which should be normalized to a common
value will not trigger the appropriate treatment due to e.g. statistical
fluctuations in each run's calculation of a cross-section used in the
normalization. And anything more complex than a global scaling (e.g. calculation
of a ratio or asymmetry) cannot be handled at all with a post-hoc scaling
treatment. The --assume-normalized command line option will force all histograms
to be treated as if they are normalized in the input, which can be useful if
you know that all the output histograms are indeed of this nature. If they are
not, it will go wrong: you have been warned!
Please use this script as a template if you need to do something more specific.
NOTE: there are many possible desired behaviours when merging runs, depending on
the factors above as well as whether the files being merged are of homogeneous
type, heterogeneous type, or a combination of both. It is tempting, therefore,
to add a large number of optional command-line parameters to this script, to
handle these cases. Experience from Rivet 1.x suggests that this is a bad idea:
if a problem is of programmatic complexity then a command-line interface which
attempts to solve it in general is doomed to both failure and unusability. Hence
we will NOT add extra arguments for applying different merging weights or
strategies based on analysis object path regexes, auto-identifying 'types' of
run, etc., etc.: if you need to merge data files in such complex ways, please
use this script as a template around which to write logic that satisfies your
particular requirements.
| yodamerge_tmp.action |
Definition at line 116 of file yodamerge_tmp.py.
| dict yodamerge_tmp.analysisobjects_in = {} |
Definition at line 136 of file yodamerge_tmp.py.
| dict yodamerge_tmp.analysisobjects_out = {} |
Definition at line 150 of file yodamerge_tmp.py.
| yodamerge_tmp.ao_out = aos[0].clone() |
Definition at line 219 of file yodamerge_tmp.py.
| list yodamerge_tmp.aos = yoda.read(filename, patterns=args.PATHPATTERNS, unpatterns=args.PATHUNPATTERNS) |
Definition at line 145 of file yodamerge_tmp.py.
Definition at line 185 of file yodamerge_tmp.py.
Definition at line 154 of file yodamerge_tmp.py.
| yodamerge_tmp.args = parser.parse_args() |
Definition at line 126 of file yodamerge_tmp.py.
Definition at line 241 of file yodamerge_tmp.py.
| yodamerge_tmp.default |
Definition at line 106 of file yodamerge_tmp.py.
| yodamerge_tmp.dest |
Definition at line 106 of file yodamerge_tmp.py.
| yodamerge_tmp.dim = ao_out.dim() |
Definition at line 239 of file yodamerge_tmp.py.
| dict yodamerge_tmp.em_i = {} |
Definition at line 254 of file yodamerge_tmp.py.
| dict yodamerge_tmp.ep_i = {} |
Definition at line 253 of file yodamerge_tmp.py.
| yodamerge_tmp.False |
Definition at line 116 of file yodamerge_tmp.py.
| yodamerge_tmp.filename |
Definition at line 138 of file yodamerge_tmp.py.
| yodamerge_tmp.help |
Definition at line 105 of file yodamerge_tmp.py.
| yodamerge_tmp.metavar |
Definition at line 106 of file yodamerge_tmp.py.
Definition at line 158 of file yodamerge_tmp.py.
| yodamerge_tmp.nargs |
Definition at line 105 of file yodamerge_tmp.py.
| yodamerge_tmp.npoints = len(ao_out.points()) |
Definition at line 250 of file yodamerge_tmp.py.
| yodamerge_tmp.parser = argparse.ArgumentParser(usage=__doc__) |
Definition at line 104 of file yodamerge_tmp.py.
| float yodamerge_tmp.rescale = None |
Definition at line 198 of file yodamerge_tmp.py.
| yodamerge_tmp.S1D_MODE |
Definition at line 131 of file yodamerge_tmp.py.
| yodamerge_tmp.S2D_MODE |
Definition at line 132 of file yodamerge_tmp.py.
| yodamerge_tmp.S3D_MODE |
Definition at line 133 of file yodamerge_tmp.py.
| yodamerge_tmp.sao = ao.mkScatter() |
Definition at line 163 of file yodamerge_tmp.py.
| list yodamerge_tmp.saos = [] |
Definition at line 161 of file yodamerge_tmp.py.
Definition at line 166 of file yodamerge_tmp.py.
| yodamerge_tmp.scale = float(scale) |
Definition at line 138 of file yodamerge_tmp.py.
| bool yodamerge_tmp.scatter_fail = False |
Definition at line 159 of file yodamerge_tmp.py.
Definition at line 240 of file yodamerge_tmp.py.
| yodamerge_tmp.True |
Definition at line 118 of file yodamerge_tmp.py.
| float yodamerge_tmp.val_i = 0.0 |
Definition at line 252 of file yodamerge_tmp.py.
| yodamerge_tmp.variations = ao.variations() |
Definition at line 257 of file yodamerge_tmp.py.
1.14.0