Optimizee for MNIST

The fitness function to optimize is the MNIST performance.

MNISTOptimizee

class l2l.optimizees.mnist.optimizee.MNISTOptimizee(traj, parameters)[source]

Bases: Optimizee

Implements a simple function optimizee. Functions are generated using the FunctionGenerator. NOTE: Make sure the optimizee_fitness_weights is set to (-1,) to minimize the value of the function

Parameters:

traj – The trajectory used to conduct the optimization.
parameters – Instance of namedtuple() MNISTOptimizeeParameters

create_individual()[source]: Creates a random value of parameter within given bounds

bounding_func(individual)[source]: Bounds the individual within the required bounds via coordinate clipping

simulate(traj)[source]

Returns the value of the function chosen during initialization

Parameters:: traj (Trajectory) – Trajectory
Returns:: a single element tuple containing the value of the chosen function

MNISTOptimizeeParameters

class l2l.optimizees.mnist.optimizee.MNISTOptimizeeParameters(n_hidden, seed, use_small_mnist)

n_hidden

seed

use_small_mnist