API Usage

Badger can be imported as a regular python package, and you could use the plugins/utils that Badger offers in your own python script.

Heads-up

Make sure you have Badger installed and setup.

Use an algorithm

Badger has a get_algo API to get a specific algorithm.

The following code gets an algorithm named silly (which is a random search algorithm) from Badger.

import numpy as np
from badger.factory import get_algo

# Define a test evaluate function
def evaluate(X):
    Y = np.linalg.norm(X, axis=1).reshape(-1, 1)  # objectives
    I = None  # inequality constraints
    E = None  # equality constraints

    # Show the progress
    print(Y)

    return Y, I, E

# Get the silly algorithm from Badger
optimize, configs = get_algo('silly')

# Optimize the test evaluate function
optimize(evaluate, configs['params'])

Use an interface

Badger has a get_intf API to get a specific interface.

The following code gets an interface named silly and constructs an instance of the interface.

from badger.factory import get_intf

# Get the silly interface from Badger
Interface, configs = get_intf('silly')
intf = Interface(configs['params'])

# Test get/set channels
intf.get_value('c1')
# Output: 0

intf.set_value('c1', 1.0)
intf.get_value('c1')
# Output: 1.0

Use an environment

Badger has a get_env API to get a specific environment.

The following code gets and instantiates an environment named silly from Badger. Note that it uses the silly interface instance intf from the last section.

from badger.factory import get_env

# Get the silly environment from Badger
Environment, configs = get_env('silly')
env = Environment(intf, configs['params'])

# Investigate the silly env
env.list_vars()
# Output: ['q1', 'q2', 'q3', 'q4']

env.list_obses()
# Output: ['l1', 'l2']

env.get_var('q1')  # q1 in env maps to c1 in intf
# Output: 1.0

env.get_obs('l2')  # l2 norm of (q1, q2, q3, q4)
# Output: 1.0

env.set_var('q2', 1)
env.get_obs('l2')
# Output: 1.4142135623730951

Now we can define an evaluate function based on the silly env, and use the silly algorithm from the use an algorithm section to optimize it.

# Define an evaluate function based on the env
def evaluate(X):
    # Note that X is a 2D array
    Y = []
    for x in X:
        env.set_vars(['q1', 'q2', 'q3', 'q4'][:len(x)], x)
        y = env.get_obs('l2')
        Y.append(y)
    Y = np.array(Y).reshape(-1, 1)
    I = None
    E = None

    # Show the progress
    print(Y)

    return Y, I, E

# Optimize the evaluate function with silly algorithm
optimize(evaluate, {'dimension': 4, 'max_iter': 42})