Paropt - generic command-line parameter optimization

Paropt is a python tool for optimizing an application's parameters using system calls and parsing performance data from STDOUT. From a starting value, it will call the command line given to it repeatedly, converging into optimal parameters using the Nelder-mead algorithm or grid search.

Usage

 popt [options] {min,max} fn_call fval_regex x0 x1 x2 ...
 popt -g [options] {min,max} fn_call fval_regex x0range x1range x2range ...

Positional Arguments

• {min,max} The direction of optimization
• fn_call A valid system call with placeholders ({} (only Python 2.7 or higher) or numbered with {0}, {1}, etc.) representing parameters, e.g. 'examples/rosenbrock.py {} {}'.
• fval_regex A valid regular expression that contains a capturing group representing the function value to parse out, e.g. 'fx = (.*)'

All additional positional arguments will represent initial values for the variables - make sure the number of variables matches your function call!

Optional Arguments

• -g --grid Use grid search (see Range specifications for details on how to specify value ranges)
• -n --neldermead Use nelder-mead (default)
• -h --help Display help
• -t --xtol Specify a tolerance value for the optimization (default: 1e-6)
• -i --maxiter Specify the maximum number of iterations for the optimization (default: None)

Range specifications

If optimizing using grid search, variables are not given as initial values but as ranges specifying the grid search space. Ranges are specified as 1-, 2- or 3-tuples giving minimal and maximal values plus an optional increment. Examples:

0,1,0.1	A range from 0 to 1 in increments of 0.1 (e.g 0.1, 0.2, ... 0.9)
1,4	A range from 1 to 4 in increments of 1 (e.g. 1, 2, 3)
0.5	A range from 0.5 to 0.5 (i.e. a constant value)

Valid delimiter characters are: ,;|

You can optionally specify lists of values for a variable by prefixing the range specification with the l character. For example, to specify the list of prime numbers smaller than 10: l2,3,5,7

Example

Nelder-mead optimization

The following example minimizes the Rosenbrock banana function located in examples/rosenbrock.py in 5 variables.

Let's say we have a python program calculating the value of the Rosenbrock function that takes values from the command line and returns the value of the function in STDOUT:

$ examples/rosenbrock.py 1.3 0.7 0.8 1.9 1.2
fx = 8.4822000000e+02 

We can minimize this function using popt:

$ ./popt --xtol 1e-8 min 'examples/rosenbrock.py {0} {1} {2} {3} {4}' 'fx = (.*)' 1.3 0.7 0.8 1.9 1.2
[...]
success: True
status: 0
nfev: 571
fun: 4.8611534334e-17
nit: 339
message: 'Optimization terminated successfully.'
x: array([ 1.,  1.,  1.,  1.,  1.])

Grid search optimization

The following example minimizes the Rosenbrock banana function located in examples/rosenbrock.py in 2 variables using grid search covering the grid [0, 0.1, ... 1.9] for each of the variables.

$ ./popt -g min 'examples/rosenbrock.py {1} {2}' 'fx = (.*)' 0,2,0.1 0,2,0.1

Dependencies

• Python 2.6 or later (including Python 3)
• scipy 0.12