Utility routines

There are a number of utility routines provided by Sherpa that may be useful. Unfortunately it is not always obvious whether a routine is for use with the Object-Oriented API or the Session API.

The Sherpa resource file

The sherpa.get_config routine returns the location of the default resource file used to configure Sherpa. The location of this file is taken from (in order):

  1. the SHERPARC environment file,

  2. the file .sherpa.rc in the user’s home directory,

  3. and the file returned by get_config.

Taking advantage of multiple CPUs

The sherpa.utils.parallel module provides code to run code on multiple CPUs. There are two settings in the Sherpa resource file that can be changed:

  • parallel.numcores

  • multiprocessing.multiprocessing_start_method

If parallel.numcores is set to 1 then the code will run in serial and not parallel.

The multiprocessing.multiprocessing_start_method field controls how the parallel code is run. Using methods other than “fork” are unlikely to provide any benefit for Sherpa, and it is probably better to set parallel.numcores to 1 if the “fork” method can not be used. The Python documentation on multiprocessing explains the methods that Python offers and discusses some pitfalls.

Sherpa relies on mutable state, in particular for handling parameter values, and so is unlikely to benefit from using threads.

Controlling the verbosity of Sherpa

Sherpa uses Python logging for most messages. This allows the user to redirected the output to a file or suppress it by setting the logging level. The following example will globally change the level for all sherpa modules, such that debug and informational messages are no longer displayed:

>>> import logging
>>> sherpalog = logging.getLogger('sherpa')
>>> sherpalog.setLevel('WARNING')

Sherpa also provides a context manager - SherpaVerbosity - to change the logging level only for a specific portion of the code. This can be used, e.g., to hide the long default output printed after fitting a model:

>>> from sherpa.utils.logging import SherpaVerbosity
>>> import numpy as np
>>> from sherpa.astro import ui
>>> ui.load_arrays("mydata", np.arange(5), np.ones(5))
>>> ui.set_model("mydata", "polynom1d.poly")
>>> with SherpaVerbosity('WARNING'):
...    ui.fit("mydata")

Random Numbers

Sherpa uses random numbers

The primary way to control the random numbers used by Sherpa is to create a NumPy Random Generator and pass it to routines either with the rng parameter or, for users of the Session class, the set_rng method. However, for the optimiser code, which uses a C++ implementation of the Mersenne Twister random number generator, the important value is the seed argument (although this can be derived from the random number generator if the seed value is set to None).

Note

Prior to Sherpa 4.16.0 the random numbers were controlled by a combination of the legacy NumPy random-number API, that is, calling numpy.random.seed() , and - in a few places, the Python random module, as well as the seed argument for the optimiser code.

Reference/API