Runana - Run programs and analyse their results

Utility library for running programs and analysing their results.

Useful for convergence testing. Integrates well with Fortran programs.

Documentation: http://runana.readthedocs.org/en/latest/

Installation

runana can be installed from pypi:

$ pip install runana

The latest version of runana can be installed from source:

$ git clone https://github.com/jenssss/runana.git
$ cd runana
$ python setup.py install

Users without install privileges can append the --user flag to setup.py:

$ python setup.py install --user

Example usage

A number of examples are included in the examples directory of the source code. The subfolder f90nml uses configuration files in the fortran namelist format, while upname uses a configuration name format in which the names and values of variables are given on consequtive lines with entries seperated by white space.

Here follows one of the examples from the f90nml folder. A simple program for performing a numerical integration is given in examples/f90nml/integrate_test.py. The main content of this file is:

#!/usr/bin/env python
from sys import argv
import numpy as np
import f90nml

config = f90nml.read(argv[1])

npoints = config['nlIntegrate']['npoints']

x = np.linspace(0, 2, npoints)
y = 10*x**2

I = np.trapz(y, x)

print('Integral of 10*x**2 from 0 to 2: ', I)

The program can be configured through a namelist configuration, which should be given as the first argument when calling the program ./intergrate_test.py config.nml. An example of such a configuration is located at examples/f90nml/config.nml and contains entries of the form:

&nlGroup
  var = 1
&end

&nlIntegrate
  npoints = 10
&end

We want to run this program for a number of different values of the npoints parameter, and compare the results. For this we can use runana. The file examples/f90nml/run_integrate.py contains a script showing how this can be run:

from os import path, getcwd
from runana.run import execute, print_time, generate_list

def setup_programs():
    programs = ['integrate_test.py',]
    programs = [path.join(getcwd(), program) for program in programs]
    return programs

def setup_replacers():
    nvar_values = 10
    chain_iters = {('nlIntegrate', 'npoints'): generate_list(
        start=10, incr=10, incr_func='add', nvalues=nvar_values),
    }
    return chain_iters

input_file = 'config.nml'

chain_iters = setup_replacers()

scratch_base = path.expanduser('~/test_run/runana/integrate_test')

programs = setup_programs()

print('Running in ', scratch_base)

with print_time():
    execute(programs, input_file, scratch_base,
            chain_iters=chain_iters)

Running this script will run the integration program with 10 values of the npoints parameter in increments of 10 starting from 10. The results of the calculations will be stored in ~/test_run/runana/integrate_test, specified in the scratch_base variable. For each parameter, a seperate run of the program will be performed, and the results stored in separate subdirectories of ~/test_run/runana/integrate_test. This script can be run by running python run_integrate.py in the examples/f90nml/ directory.

Finally, the results can be analyzed using the script in examples/f90nml/analyse_integrate.py, which contains:

from os import path
from runana import analyse
from runana import analyse_pandas
from runana import read_numbers

workdir = path.expanduser('~/test_run/runana/integrate_test')

params_to_dirs = analyse.read_input_files(workdir)

params_to_dirs.diff()

panda_data = analyse_pandas.make_a_seq_panda(params_to_dirs)

read_var = analyse.make_collector_function(
      workdir,
      read_numbers.read_last_number_from_file,
      fname="integrate_test.py.stdout",
      pattern="Integral",
  )
panda_var = panda_data.applymap(read_var)
print("Values of integral")
print(panda_var)

panda_conv = panda_var.calc_convergence()
print("Estimated difference between current and fully converged value")
print(panda_conv)
param_panda = panda_data.applymap(
      analyse_pandas.return_dict_element(params_to_dirs)
      )
panda_var.plot_("plot_test_integral_var.pdf", param_panda=param_panda)
panda_conv.plot_("plot_test_integral_conv.pdf", logy=True, param_panda=param_panda)

Running this script should print out:

Values of integral:
                              0
NumParam NumParamValue
npoints  10.0           26.831276
         20.0           26.703601
         30.0           26.682521
         40.0           26.675433
         50.0           26.672220
         60.0           26.670497
         70.0           26.669467
         80.0           26.668803
         90.0           26.668350
         100.0          26.668027

Estimated difference between current and fully converged value:
                          0_conv
NumParam NumParamValue
npoints  10.0                NaN
         20.0           0.009562
         30.0           0.002370
         40.0           0.001063
         50.0           0.000602
         60.0           0.000388
         70.0           0.000270
         80.0           0.000199
         90.0           0.000153
         100.0          0.000121

The script collects the values calculated by the integration program and puts them into a pandas DataFrame, indexed by the value of the varying numerical parameter. It also calculates an estimate for how well converged the calculation is. Finally the script plots these values to the files plot_test_integral_var.pdf and plot_test_integral_conv.pdf.

Indices and tables