3. File Input/Output

The primary package is scisoftpy.io.

3.1. Loading

Images and recorded data from various detectors can be read into the data analysis plugin. This is done using the following method:

import scisoftpy as dnp
dh = dnp.io.load("datafile")

This loads up the file as a list/ordered dictionary of datasets. Optionally arguments can be specified:

dh = dnp.io.load(name, format=None, withmetadata=True, ascolour=False, warn=True)


name is a file name

format is a list of strings that specify the formats to try

  • png, gif, jpeg, tiff, pgm - standard image formats
  • adsc - ADSC Quantum area series detector format
  • crysalis - Oxford Diffraction CrysAlis processing software format
  • mar - Rayonix’s MarCCD detector
  • pilatus - Dectris Pilatus detector version of TIFF
  • cbf - Crystallographic Binary Format (IUCR)
  • xmap - XIA’s DXP-xMAP format for x-ray spectra
  • srs - Daresbury Laboratory’s Synchrotron Radiation Source format
  • dls - Diamond Light Source text format
  • edf - European Synchrotron Radiation Source data format for Pilatus images
  • binary - raw single dataset binary format
  • text - raw single dataset ASCII format
  • npy - NumPy binary format
  • hdf5 - HDF5 [1] tree format
  • nx - NeXus [2] tree format

or None (default) to attempt all above formats

withmetadata is a boolean value to dictate whether to return metadata in addition to datasets

ascolour is a boolean value to dictate whether to return any coloured image data as an RGB dataset or convert it to greyscale

warn is a boolean value to dictate whether to print some warnings

The object returned by load is usually a data holder and acts like a list and ordered dictionary. It has at least one attribute metadata which is (possibly empty) list/dictionary. Values held in a data holder can be accessed by strings or integers. They can also be accessed as attributes. For example, a data file has a single dataset named ‘voltage’, then that dataset can be accessed as follows:

dh = dnp.io.load('measure.dat')
len(dh)        # returns 1
dh.keys()      # shows list of with string 'voltage'
dh[0]          # shows array of voltage values
dh['voltage']  # same as above
dh.voltage     # same as above

If the data file has a metadata item that is the temperature, this can be accessed in a similar way:

dh.metadata.keys()          # shows string 'temperature' in a list
dh.metadata[0]              # shows its value
dh.metadata['temperature']  # shows its value
dh.metadata.temperature     # shows its value

Note, the key names in a list/dictionary are transformed to valid python variable names by substitution of underscore characters for illegal characters and prepending an underscore if the name starts with a digit.

Currently, some loaders transform metadata items so that their names follow a NeXus [2] naming convention.

The four standard image loaders will load and convert RGB images to grey-scale (luma).

In some of the formats supported (CBF, MarCCD, ADSC), information is supplied on the position and orientation of the detector that took the image, the size of pixels, the position and wavelength of an illuminating beam in a diffraction experiment. This diffraction data can be loaded in exactly the same way:

di = dnp.io.load(name)

which returns a list (or dictionary) of diffraction images. These images can be plotted using the plot package.

Sometimes, warnings are shown from loading files. There is a global setting to turn these off if they become too distracting:


This setting can be overridden per load call using the warn keyword.

Loaders for the two tree formats (hdf5 and nx) return data structures that look like trees with a root node, branches to other nodes. Nodes can have branches, i.e. it is a group, and also contain node attributes and datasets. Node attributes within a node are collated in a list/dictionary under the attribute attrs. Group nodes and datasets are held in a list/dictionary on the node itself. Datasets in trees can also hold attributes. It is important to note that these datasets are lazy, see Lazy dataset.

Here is an example of loading a tree format file:

t = dnp.io.load('bl20-20237.nxs') # load tree

t.keys()       # returns the names of the nodes in the root group node
t.attrs.keys() # lists the names of the attributes

t['entry1'] # retrieves the sub-node called `entry1`
t.entry1    # same as above
t[0]        # same as above if `entry1` is the first item in the root group

g = t['entry1/data']    # retrieves the sub-sub-node (a group node)
la = g.data             # retrieves the lazy dataset with that sub-sub-sub-node
lb = t.entry1.data.data # same as above

b = lb[...] # loads entire dataset

The tree object or root group also provides a method to find all nodes of a given name:

getnodes(name, group=True, data=True)

which returns a list of nodes. By default, both groups and datasets are included in the list but this behaviour can be modified by setting the two keyword arguments: group and data. Thus:

t.getnodes('data', group=False) # returns datasets named "data"

3.2. Saving

Datasets can be saved:

dnp.io.save(name, data, format=None, range=(), autoscale=False, signed=True, bits=None)


name is a file name

data is a dataset or sequence of datasets

format is one of following strings

  • png, gif, jpeg, tiff - standard image formats
  • text - raw ASCII dump
  • binary - raw binary dump
  • npy - NumPy binary format

or None (default) to guess format from file name extension

range is a tuple for minimum and maximum values for clipping a dataset
before saving
autoscale is a boolean value to dictate whether to scale automatically
dataset values to fit the chosen format (only png and jpeg are supported for auto-scaling).

signed is a boolean value to dictate whether to save as signed numbers

bits is an integer value to dictate the number of bits to use when saving

If there are multiple datasets specified then multiple images will be saved, suffixed with a number representing the number of the dataset in the sequence. The bits keyword is supported by png and tiff savers. The latter saver supports the unsigned keyword too. If the number of bits used is greater than 32, the saver attempts to use 32-bit floats.