Skip to main content

python interface to the hyperparameter optimization tool SMAC.

Project description

Simple python wrapper to SMAC, a versatile tool for optimizing algorithm parameters.

fmin(objective, x0, xmin, xmax, x0_int, xmin_int, xmax_int, xcategorical, params)
   min_x f(x) s.t. xmin < x < xmax

 objective: The objective function that should be optimized.



pip install pysmac


python install

Example usage

Let’s take for example the Branin function. (Note that the branin function is not the ideal use case for SMAC, which is designed to be a global optimization tool for costly functions. That said, it’ll serve the purpose of checking that everything is working.)

import numpy as np

def branin(x):
    b = (5.1 / (4.*np.pi**2))
    c = (5. / np.pi)
    t = (1. / (8.*np.pi))
    return 1.*(x[1]-b*x[0]**2+c*x[0]-6.)**2+10.*(1-t)*np.cos(x[0])+10.

For x1 ∈ [-5, 10], x2 ∈ [0, 15] the function reaches a minimum value of: 0.397887.

Note: fmin accepts any function that has a parameter called x (the input array) and returns an objective value.

from pysmac.optimize import fmin

xmin, fval = fmin(branin, x0=(0,0),xmin=(-5, 0), xmax=(10, 15), max_evaluations=5000)

As soon as the evaluations are finished, we can check the output:

>>> xmin
{'x': array([ 3.14305644,  2.27827543])}

>>> fval

Let’s run the objective function with the found parameters:

>>> branin(**xmin)


SMAC is free for academic & non-commercial usage. Please contact Frank Hutter to discuss obtaining a license for commercial purposes.


Custom arguments to the objective function:

Note: make sure there is no naming collission with the parameter names and the custom arguments.

def minfunc(x, custom_arg1, custom_arg2):
    print "custom_arg1:", custom_arg1
    print "custom_arg2:", custom_arg2
    return 1

xmin, fval = fmin(minfunc, x0=(0,0),xmin=(-5, 0), xmax=(10, 15),
                  custom_args={"custom_arg1": "test",
                               "custom_arg2": 123})

Integer parameters

Integer parameters can be encoded as follows:

def minfunc(x, x_int):
    print "x: ", x
    print "x_int: ", x_int
    return 1.

xmin, fval = fmin(minfunc,
                  x0=(0,0), xmin=(-5, 0), xmax=(10, 15),
                  x0_int=(0,0), xmin_int=(-5, 0), xmax_int=(10, 15),

Categorical parameters

Categorical parameters can be specified as a dictionary of lists of values they can take on, e.g.:

categorical_params = {"param1": [1,2,3,4,5,6,7],
                      "param2": ["string1", "string2", "string3"]}

def minfunc(x_categorical):
    print "param1: ", x_categorical["param1"]
    print "param2: ", x_categorical["param2"]
    return 1.

xmin, fval = fmin(minfunc,

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pysmac-0.8.tar.gz (9.5 MB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page