Config.py

from GlobalDefines import *


#default settings (for possible options see GlobalDefines.py)
DEFAULT_SELECTION_POLICY = GLODEF_SELECTION_UCBTUNED
    #GLODEF_SELECTION_RANDOM = 0
    #GLODEF_SELECTION_EGREEDY = 1
    #GLODEF_SELECTION_SOFTMAX = 2
    #GLODEF_SELECTION_UCB1 = 3
    #GLODEF_SELECTION_UCBTUNED = 4

DEFAULT_CHANGEPOINT_DETECTOR = GLODEF_CHANGEPOINT_HENKYPENKY
    #GLODEF_CHANGEPOINT_NONE = 0
    #GLODEF_CHANGEPOINT_Z_TEST = 1
    #GLODEF_CHANGEPOINT_HENKYPENKY = 2

DEFAULT_CHANGEPOINT_TEST = GLODEF_CHANGEPOINT_TEST_Z1
    #GLODEF_CHANGEPOINT_TEST_Z1 = 0
    #GLODEF_CHANGEPOINT_TEST_Z2 = 1

DEFAULT_RESET_ALGORITHM = GLODEF_RESET_ALGORITHM_RESET_TO_MOVING_AVERAGE
    #GLODEF_RESET_ALGORITHM_RESET_ALL_TO_ZERO = 0
    #GLODEF_RESET_ALGORITHM_RESET_ALL_TO_MOVING_AVERAGE = 1
    #GLODEF_RESET_ALGORITHM_RESET_ALL_TO_MOVING_AVERAGE_CUTOFF = 2
    #GLODEF_RESET_ALGORITHM_RESET_TO_MOVING_AVERAGE = 3

DEFAULT_PARAM_FUNCTIONS = GLODEF_PARAM_FUNCTION_DIRECT
    #GLODEF_PARAM_FUNCTION_DIRECT = 0
    #GLODEF_PARAM_FUNCTION_LINEAR = 1
    #GLODEF_PARAM_FUNCTION_NEURAL = 2
DEFAULT_SOLVER_NUMPARAMS = 5
    # one parameter for selection algorithm, four parameters for change point detector
DEFAULT_PARAM_NUMINPUTS = 0
    # number of parameter function internal weights (stored values) are number of inputs + 1

DEFAULT_OPTIMIZATION_ALGORITHM = GLODEF_OPTIMIZATION_ANNEALING
    #GLODEF_OPTIMIZATION_ANNEALING = 0
    #GLODEF_OPTIMIZATION_GENETIC = 1
    #GLODEF_OPTIMIZATION_EXHAUSTIVE = 2
    #GLODEF_OPTIMIZATION_RANDOM = 3
DEFAULT_OPTIMIZATION_CONFIG = []
DEFAULT_OPTIMIZATION_EVALS_PER_SAMPLE = 10

DEFAULT_EVAL_NUM_METRICS = 4
DEFAULT_FITNESS_METRIC = GLODEF_FITNESS_SUMREWARDS
    #GLODEF_FITNESS_SUMREWARDS = 0
    #GLODEF_FITNESS_REGRET = 1
    #GLODEF_FITNESS_OPTIMALITY = 2
    #GLODEF_FITNESS_OPTIMALITY_RANDNOR = 3

DEFAULT_PAR_EGREEDY_E = 0.1         # epsilon-greedy parameter: Epsilon
DEFAULT_PAR_SOFTMAX_T = 0.01        # softMax parameter: tau
DEFAULT_PAR_UCB1_C = 1.0            # UCB parameter: C
DEFAULT_PAR_POKER_H = 0             # POKER parameter: H (horizon = number of rounds that remains to be played)
DEFAULT_PAR_VOTER = [1.0, 1.0, 1.0]
DEFAULT_PAR_CHANGEPOINT_THR = 2.5   # change point detector parameter: Z-Threshold
DEFAULT_PAR_CHANGEPOINT_INT = 1.0   # change point detector parameter: interval shrink (e.g., 0.6 translates interval [0,1] to [0.2,0.8]
DEFAULT_PAR_CHANGEPOINT_NUM = 50    # change point detector parameter: lowest number of moving average samples that may trigger reset
DEFAULT_PAR_CHANGEPOINT_SOFT = 0.0  # soft reset: what ammount of previous value is retained (a weighted sum of the previous and new value is taken), in range [0,1], default 0