I/O functions for Python and LQCD file formats
Project description
I/O functions for Python and LQCD file formats
Lyncs IO offers two high-level functions load
and save
(or dump
as alias of save
).
The main features of this module are
-
Seamlessly IO, reading and writing made simple. In most of the cases, after saving
save(obj, filename)
, loadingobj=load(filename)
returns the original Python object. This feature is already ensured by formats likepickle
, but we try to ensure it as much as possible also for other formats. -
Many formats supported. The file format can be specified either via the filename's extension or with the option
format
passed toload/save
. The structure of the package is flexible enough to easily accomodate new/customized file formats as these arise. See [Adding a file format] for guidelines. -
Support for archives. In case of archives, e.g. HDF5, zip etc., the content can be accessed directly by specifying it in the path. For instance with
directory/file.h5/content
,directory/file.h5
is the file path, and the remaining is content to be accessed that will be searched inside the file. -
Support for Parallel IO. Where possible, the option
chunks
can be used for enabling parallel IO viaDask
. -
Omission of extension. When saving, if the extension is omitted, the optimal file format is deduced from the data type and the extension is added to the filename. When loading, any extension is considered, i.e.
filename.*
, and if only one match is available, the file is loaded.
Installation
The package can be installed via pip
:
pip install [--user] lyncs_io
NOTE: for enabling parallel IO, lyncs_io requires a working MPI installation.
This can be installed via apt-get
:
sudo apt-get install libopenmpi-dev openmpi-bin
OR using conda
:
conda install -c anaconda mpi4py
Parallel IO can then be enabled via
pip install [--user] lyncs_io[mpi]
Documentation
The high-level load
and save
(or dump
as alias of save
) functions provided by the Lyncs IO can be used as follows:
import numpy as np
import lyncs_io as io
arr1 = np.random.rand((10,10,10))
io.save(arr, "data.h5/random")
arr2 = np.zeros_like(arr)
io.save(arr, "data.h5/zeros")
arrs = io.load("data.h5")
assert (arr1 == arrs["random"]).all()
assert (arr2 == arrs["zeros"]).all()
NOTE: for save
we use the order data, filename
. This is the opposite
of what done in numpy
but consistent with pickle
's dump
. This order
is preferred because the function can be used directly as a method
for a class since self
, i.e. the data
, would be passed as the first
argument of save
.
IO with MPI
import numpy as np
import lyncs_io as io
from mpi4py import MPI
# Assume 2D cartesian topology
comm = MPI.COMM_WORLD
dims = MPI.Compute_dims(comm.size, 2)
cartesian2d = comm.Create_cart(dims=dims)
oarr = np.random.rand(6, 4, 2, 2)
io.save(oarr, "pario.npy", comm=cartesian2d)
iarr = io.load("pario.npy", comm=cartesian2d)
assert (iarr == oarr).all()
NOTE: Parallel IO is enabled once a valid cartesian communicator is passed to load
or save
routines, otherwise Serial IO is performed. Currently only numpy
format supports this functionality.
File formats
Adding a file format
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.