pdmlabs.mango

class pdmlabs.mango.MetaTuner(param_dict_list, objective_list, **kwargs)

Bases: object

class Config(n_iter: int = 20, n_init: int = 2)

Bases: object

n_init: int = 2

n_iter: int = 20

static calculateDomainSize(param_dict)

Calculating the domain size to be explored for finding optimum of bayesian optimizer

get_max_y_value(Y_dict_array)

run()

runExponentialTuner()

Steps: 1-Create DS obj for each obj of param_dict_list 2-Sample randomly from the DS objects and evaluate the objective functions. 3- Now use the GPR for each objective to select next batch 4- Select the best values based on fxn(surrogate) from GPR. 5- Modify the surrogate selections so as to avoid getting struck.

class pdmlabs.mango.Tuner(param_dict, objective, conf_dict=None)

Bases: object

class Config(domain_size: int = None, initial_random: int = 2, initial_custom: list = None, num_iteration: int = 20, batch_size: int = 1, optimizer: str = 'Bayesian', parallel_strategy: str = 'clustering', surrogate: object = None, alpha: float = 2.0, exploration: float = 1.0, exploration_decay: float = 0.9, exploration_min: float = 0.1, fixed_domain: bool = False, early_stopping: Callable = None, constraint: Callable = None)

Bases: object

alpha: float = 2.0

batch_size: int = 1

constraint: Callable = None

domain_size: int = None

early_stop(results)

early_stopping: Callable = None

exploration: float = 1.0

exploration_decay: float = 0.9

exploration_min: float = 0.1

fixed_domain: bool = False

initial_custom: list = None

initial_random: int = 2

property is_bayesian

property is_random

num_iteration: int = 20

optimizer: str = 'Bayesian'

parallel_strategy: str = 'clustering'

property strategy_is_clustering

property strategy_is_penalty

surrogate: object = None

valid_optimizers = ['Bayesian', 'Random']

valid_parallel_strategies = ['penalty', 'clustering']

static calculateDomainSize(param_dict)

Calculating the domain size to be explored for finding optimum of bayesian optimizer

maximize()

minimize()

process_initial_custom()

run()

runBayesianOptimizer()

runRandomOptimizer()

runUserObjective(X_next_PS)

run_initial()

Modules

domain
metatuner	Meta Tuner Class: Used to optimize across a set of models: - selecting intelligently the order of functions to optimize Current implementation: Bare Metal functionality for testing.
optimizer
scheduler
tuner	Main Tuner Class which uses other abstractions.