Utilities¶
PyTables comes with a couple of utilities that make the life easier to the user. One is called ptdump and lets you see the contents of a PyTables file (or generic HDF5 file, if supported). The other one is named ptrepack that allows to (recursively) copy sub-hierarchies of objects present in a file into another one, changing, if desired, some of the filters applied to the leaves during the copy process.
Normally, these utilities will be installed somewhere in your PATH during the process of installation of the PyTables package, so that you can invoke them from any place in your file system after the installation has successfully finished.
ptdump¶
As has been said before, ptdump utility allows you look into the contents of your PyTables files. It lets you see not only the data but also the metadata (that is, the structure and additional information in the form of attributes).
Usage¶
For instructions on how to use it, just pass the -h flag to the command:
$ ptdump -h
to see the message usage:
usage: ptdump [-d] [-v] [-a] [-c] [-i] [-R start,stop,step] [-h] file[:nodepath]
-d -- Dump data information on leaves
-v -- Dump more metainformation on nodes
-a -- Show attributes in nodes (only useful when -v or -d are active)
-c -- Show info of columns in tables (only useful when -v or -d are active)
-i -- Show info of indexed columns (only useful when -v or -d are active)
-R RANGE -- Select a RANGE of rows in the form "start,stop,step"
-h -- Print help on usage
Read on for a brief introduction to this utility.
A small tutorial on ptdump¶
Let’s suppose that we want to know only the structure of a file. In order to do that, just don’t pass any flag, just the file as parameter.
$ ptdump vlarray1.h5
/ (RootGroup) ''
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
/vlarray2 (VLArray(3,), shuffle, zlib(1)) 'ragged array of strings'
we can see that the file contains just a leaf object called vlarray1, that is an instance of VLArray, has 4 rows, and two filters has been used in order to create it: shuffle and zlib (with a compression level of 1).
Let’s say we want more meta-information. Just add the -v (verbose) flag:
$ ptdump -v vlarray1.h5
/ (RootGroup) ''
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
atom = Int32Atom(shape=(), dflt=0)
byteorder = 'little'
nrows = 3
flavor = 'numpy'
/vlarray2 (VLArray(3,), shuffle, zlib(1)) 'ragged array of strings'
atom = StringAtom(itemsize=2, shape=(), dflt='')
byteorder = 'irrelevant'
nrows = 3
flavor = 'python'
so we can see more info about the atoms that are the components of the vlarray1 dataset, i.e. they are scalars of type Int32 and with NumPy flavor.
If we want information about the attributes on the nodes, we must add the -a flag:
$ ptdump -va vlarray1.h5
/ (RootGroup) ''
/._v_attrs (AttributeSet), 4 attributes:
[CLASS := 'GROUP',
PYTABLES_FORMAT_VERSION := '2.0',
TITLE := '',
VERSION := '1.0']
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
atom = Int32Atom(shape=(), dflt=0)
byteorder = 'little'
nrows = 3
flavor = 'numpy'
/vlarray1._v_attrs (AttributeSet), 3 attributes:
[CLASS := 'VLARRAY',
TITLE := 'ragged array of ints',
VERSION := '1.3']
/vlarray2 (VLArray(3,), shuffle, zlib(1)) 'ragged array of strings'
atom = StringAtom(itemsize=2, shape=(), dflt='')
byteorder = 'irrelevant'
nrows = 3
flavor = 'python'
/vlarray2._v_attrs (AttributeSet), 4 attributes:
[CLASS := 'VLARRAY',
FLAVOR := 'python',
TITLE := 'ragged array of strings',
VERSION := '1.3']
Let’s have a look at the real data:
$ ptdump -d vlarray1.h5
/ (RootGroup) ''
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
Data dump:
[0] [5 6]
[1] [5 6 7]
[2] [5 6 9 8]
/vlarray2 (VLArray(3,), shuffle, zlib(1)) 'ragged array of strings'
Data dump:
[0] ['5', '66']
[1] ['5', '6', '77']
[2] ['5', '6', '9', '88']
We see here a data dump of the 4 rows in vlarray1 object, in the form of a list. Because the object is a VLA, we see a different number of integers on each row.
Say that we are interested only on a specific row range of the /vlarray1 object:
ptdump -R2,3 -d vlarray1.h5:/vlarray1
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
Data dump:
[2] [5 6 9 8]
Here, we have specified the range of rows between 2 and 4 (the upper limit excluded, as usual in Python). See how we have selected only the /vlarray1 object for doing the dump (vlarray1.h5:/vlarray1).
Finally, you can mix several information at once:
$ ptdump -R2,3 -vad vlarray1.h5:/vlarray1
/vlarray1 (VLArray(3,), shuffle, zlib(1)) 'ragged array of ints'
atom = Int32Atom(shape=(), dflt=0)
byteorder = 'little'
nrows = 3
flavor = 'numpy'
/vlarray1._v_attrs (AttributeSet), 3 attributes:
[CLASS := 'VLARRAY',
TITLE := 'ragged array of ints',
VERSION := '1.3']
Data dump:
[2] [5 6 9 8]
ptrepack¶
This utility is a very powerful one and lets you copy any leaf, group or complete subtree into another file. During the copy process you are allowed to change the filter properties if you want so. Also, in the case of duplicated pathnames, you can decide if you want to overwrite already existing nodes on the destination file. Generally speaking, ptrepack can be useful in may situations, like replicating a subtree in another file, change the filters in objects and see how affect this to the compression degree or I/O performance, consolidating specific data in repositories or even importing generic HDF5 files and create true PyTables counterparts.
Usage¶
For instructions on how to use it, just pass the -h flag to the command:
$ ptrepack -h
to see the message usage:
usage: ptrepack [-h] [-v] [-o] [-R start,stop,step] [--non-recursive] [--dest-title=title] [--dont-create-sysattrs] [--dont-copy-userattrs] [--overwrite-nodes] [--complevel=(0-9)] [--complib=lib] [--shuffle=(0|1)] [--fletcher32=(0|1)] [--keep-source-filters] [--chunkshape=value] [--upgrade-flavors] [--dont-regenerate-old-indexes] [--sortby=column] [--checkCSI] [--propindexes] sourcefile:sourcegroup destfile:destgroup
-h -- Print usage message.
-v -- Show more information.
-o -- Overwrite destination file.
-R RANGE -- Select a RANGE of rows (in the form "start,stop,step")
during the copy of \*all* the leaves. Default values are
"None,None,1", which means a copy of all the rows.
--non-recursive -- Do not do a recursive copy. Default is to do it.
--dest-title=title -- Title for the new file (if not specified,
the source is copied).
--dont-create-sysattrs -- Do not create sys attrs (default is to do it).
--dont-copy-userattrs -- Do not copy the user attrs (default is to do it).
--overwrite-nodes -- Overwrite destination nodes if they exist. Default is
to not overwrite them.
--complevel=(0-9) -- Set a compression level (0 for no compression, which
is the default).
--complib=lib -- Set the compression library to be used during the copy.
lib can be set to "zlib", "lzo", "bzip2" or "blosc".
Additional compressors for Blosc like "blosc:blosclz",
"blosc:lz4", "blosc:lz4hc", "blosc:snappy" and
"blosc:zlib", are supported too. Defaults to "zlib".
--shuffle=(0|1) -- Activate or not the shuffling filter (default is active
if complevel>0).
--fletcher32=(0|1) -- Whether to activate or not the fletcher32 filter
(not active by default).
--keep-source-filters -- Use the original filters in source files. The
default is not doing that if any of --complevel, --complib, --shuffle
or --fletcher32 option is specified.
--chunkshape=("keep"\|"auto"\|int|tuple) -- Set a chunkshape. A value
of "auto" computes a sensible value for the chunkshape of the
leaves copied. The default is to "keep" the original value.
--upgrade-flavors -- When repacking PyTables 1.x files, the flavor of
leaves will be unset. With this, such a leaves will be serialized
as objects with the internal flavor ('numpy' for 2.x series).
--dont-regenerate-old-indexes -- Disable regenerating old indexes. The
default is to regenerate old indexes as they are found.
--sortby=column -- Do a table copy sorted by the index in "column".
For reversing the order, use a negative value in the "step" part of
"RANGE" (see "-R" flag). Only applies to table objects.
--checkCSI -- Force the check for a CSI index for the --sortby column.
--propindexes -- Propagate the indexes existing in original tables. The
default is to not propagate them. Only applies to table objects.
Read on for a brief introduction to this utility.
A small tutorial on ptrepack¶
Imagine that we have ended the tutorial 1 (see the output of examples/tutorial1-1.py), and we want to copy our reduced data (i.e. those datasets that hangs from the /column group) to another file. First, let’s remember the content of the examples/tutorial1.h5:
$ ptdump tutorial1.h5
/ (RootGroup) 'Test file'
/columns (Group) 'Pressure and Name'
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
/detector (Group) 'Detector information'
/detector/readout (Table(10,)) 'Readout example'
Now, copy the /columns to other non-existing file. That’s easy:
$ ptrepack tutorial1.h5:/columns reduced.h5
That’s all. Let’s see the contents of the newly created reduced.h5 file:
$ ptdump reduced.h5
/ (RootGroup) ''
/name (Array(3,)) 'Name column selection'
/pressure (Array(3,)) 'Pressure column selection'
so, you have copied the children of /columns group into the root of the reduced.h5 file.
Now, you suddenly realized that what you intended to do was to copy all the hierarchy, the group /columns itself included. You can do that by just specifying the destination group:
$ ptrepack tutorial1.h5:/columns reduced.h5:/columns
$ ptdump reduced.h5
/ (RootGroup) ''
/name (Array(3,)) 'Name column selection'
/pressure (Array(3,)) 'Pressure column selection'
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
OK. Much better. But you want to get rid of the existing nodes on the new file. You can achieve this by adding the -o flag:
$ ptrepack -o tutorial1.h5:/columns reduced.h5:/columns
$ ptdump reduced.h5
/ (RootGroup) ''
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
where you can see how the old contents of the reduced.h5 file has been overwritten.
You can copy just one single node in the repacking operation and change its name in destination:
$ ptrepack tutorial1.h5:/detector/readout reduced.h5:/rawdata
$ ptdump reduced.h5
/ (RootGroup) ''
/rawdata (Table(10,)) 'Readout example'
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
where the /detector/readout has been copied to /rawdata in destination.
We can change the filter properties as well:
$ ptrepack --complevel=1 tutorial1.h5:/detector/readout reduced.h5:/rawdata
Problems doing the copy from 'tutorial1.h5:/detector/readout' to 'reduced.h5:/rawdata'
The error was --> tables.exceptions.NodeError: destination group \``/\`` already has a node named \``rawdata``; you may want to use the \``overwrite`` argument
The destination file looks like:
/ (RootGroup) ''
/rawdata (Table(10,)) 'Readout example'
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
Traceback (most recent call last):
File "utils/ptrepack", line 3, in ?
main()
File ".../tables/scripts/ptrepack.py", line 349, in main
stats = stats, start = start, stop = stop, step = step)
File ".../tables/scripts/ptrepack.py", line 107, in copy_leaf
raise RuntimeError, "Please check that the node names are not
duplicated in destination, and if so, add the --overwrite-nodes flag
if desired."
RuntimeError: Please check that the node names are not duplicated in
destination, and if so, add the --overwrite-nodes flag if desired.
Ooops! We ran into problems: we forgot that the /rawdata pathname already existed in destination file. Let’s add the –overwrite-nodes, as the verbose error suggested:
$ ptrepack --overwrite-nodes --complevel=1 tutorial1.h5:/detector/readout
reduced.h5:/rawdata
$ ptdump reduced.h5
/ (RootGroup) ''
/rawdata (Table(10,), shuffle, zlib(1)) 'Readout example'
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
you can check how the filter properties has been changed for the /rawdata table. Check as the other nodes still exists.
Finally, let’s copy a slice of the readout table in origin to destination, under a new group called /slices and with the name, for example, aslice:
$ ptrepack -R1,8,3 tutorial1.h5:/detector/readout reduced.h5:/slices/aslice
$ ptdump reduced.h5
/ (RootGroup) ''
/rawdata (Table(10,), shuffle, zlib(1)) 'Readout example'
/columns (Group) ''
/columns/name (Array(3,)) 'Name column selection'
/columns/pressure (Array(3,)) 'Pressure column selection'
/slices (Group) ''
/slices/aslice (Table(3,)) 'Readout example'
note how only 3 rows of the original readout table has been copied to the new aslice destination. Note as well how the previously nonexistent slices group has been created in the same operation.
pt2to3¶
The PyTables 3.x series now follows PEP 8 coding standard. This makes using PyTables more idiomatic with surrounding Python code that also adheres to this standard. The primary way that the 2.x series was not PEP 8 compliant was with respect to variable naming conventions. Approximately 450 API variables were identified and updated for PyTables 3.x.
To ease migration, PyTables ships with a new pt2to3 command line tool.
This tool will run over a file and replace any instances of the old variable
names with the 3.x version of the name. This tool covers the overwhelming
majority of cases was used to transition the PyTables code base itself! However,
it may also accidentally also pick up variable names in 3rd party codes that
have exactly the same name as a PyTables’ variable. This is because pt2to3
was implemented using regular expressions rather than a fancier AST-based
method. By using regexes, pt2to3 works on Python and Cython code.
pt2to3 help:
usage: pt2to3 [-h] [-r] [-p] [-o OUTPUT] [-i] filename
PyTables 2.x -> 3.x API transition tool This tool displays to standard out, so
it is common to pipe this to another file: $ pt2to3 oldfile.py > newfile.py
positional arguments:
filename path to input file.
optional arguments:
-h, --help show this help message and exit
-r, --reverse reverts changes, going from 3.x -> 2.x.
-p, --no-ignore-previous
ignores previous_api() calls.
-o OUTPUT output file to write to.
-i, --inplace overwrites the file in-place.
Note that pt2to3 only works on a single file, not a a directory. However,
a simple BASH script may be written to run pt2to3 over an entire directory
and all sub-directories:
#!/bin/bash
for f in $(find .)
do
echo $f
pt2to3 $f > temp.txt
mv temp.txt $f
done
Note
pt2to3 uses the argparse module that is part of the
Python standard library since Python 2.7.
Users of Python 2.6 should install argparse separately
(e.g. via pip).
The old APIs and variable names will continue to be supported for the short term, where possible. (The major backwards incompatible changes come from the renaming of some function and method arguments and keyword arguments.) Using the 2.x APIs in the 3.x series, however, will issue warnings. The following is the release plan for the warning types:
- 3.0 - PendingDeprecationWarning
- 3.1 - DeprecationWarning
- >=3.2 - Remove warnings, previous_api(), and _past.py; keep pt2to3,
The current plan is to maintain the old APIs for at least 2 years, though this is subject to change.
