misc.optimizers package¶

Submodules¶

misc.optimizers.cem module¶

class misc.optimizers.cem.CEMOptimizer(sol_dim, max_iters, popsize, num_elites, constrains, tf_session=None, epsilon=0.001, alpha=0.25, max_resamples=10)¶

Bases: misc.optimizers.optimizer.Optimizer

A Tensorflow-compatible CEM optimizer.

Parameters:

sol_dim (int) – The dimensionality of the problem space
max_iters (int) – The maximum number of iterations to perform during optimization
popsize (int) – The number of candidate solutions to be sampled at every iteration
num_elites (int) – The number of top solutions that will be used to obtain the distribution at the next iteration.
constrains (array) – [[np.array([min v, min q]), np.array([max v, max q])], [min q/v, max q/v], [min q/sqrt(v), max q/sqrt(v)]]
tf_session (tf.Session) – (optional) Session to be used for this optimizer. Defaults to None, in which case any functions passed in cannot be tf.Tensor-valued.
epsilon (float) – A minimum variance. If the maximum variance drops below epsilon, optimization is stopped.
alpha (float) – Controls how much of the previous mean and variance is used for the next iteration. next_mean = alpha * old_mean + (1 - alpha) * elite_mean, and similarly for variance.

changeSolDim(sol_dim)¶

Change the dimension of the CEM optimisation solution.

Parameters:	sol_dim (int) – New dimension of the CEM optimisation solution.

obtain_solution(init_mean, init_var)¶

Optimizes the cost function using the provided initial candidate distribution

Parameters:	init_mean (np.ndarray) – The mean of the initial candidate distribution. init_var (np.ndarray) – The variance of the initial candidate distribution.

reset()¶: Blank function for compatibility with optimisation class framework.

setup(cost_function, tf_compatible)¶

Sets up this optimizer using a given cost function.

Parameters:	cost_function (func) – A function for computing costs over a batch of candidate solutions. tf_compatible (bool) – True if the cost function provided is tf.Tensor-valued.

Returns: None

misc.optimizers.optimizer module¶

class misc.optimizers.optimizer.Optimizer(*args, **kwargs)¶

Bases: object

Framework for Optimizer subclasses

obtain_solution(*args, **kwargs)¶: Compute optimisation problem solution

reset()¶: Function iteratively called at MPC.act()

setup(cost_function, tf_compatible)¶: Function called upon initialisation of the MPC class.

misc.optimizers.random module¶

class misc.optimizers.random.RandomOptimizer(sol_dim, popsize, tf_session, upper_bound=None, lower_bound=None)¶