config.py

''' Configuration File.
'''

##
# Learning Loss for Active Learning
NUM_TRAIN = 50000 # N
NUM_VAL   = 50000 - NUM_TRAIN
BATCH     = 128 # B
SUBSET    = 10000 # M
ADDENDUM  = 1000 # K

MARGIN = 1.0 # xi
WEIGHT = 1.0 # lambda

TRIALS = 3
CYCLES = 10

EPOCH = 200
LR = 0.1
MILESTONES = [160]
EPOCHL = 120 # After 120 epochs, stop the gradient from the loss prediction module propagated to the target model

MOMENTUM = 0.9
WDECAY = 5e-4

''' CIFAR-10 | ResNet-18 | 93.6%
NUM_TRAIN = 50000 # N
NUM_VAL   = 50000 - NUM_TRAIN
BATCH     = 128 # B
SUBSET    = NUM_TRAIN # M
ADDENDUM  = NUM_TRAIN # K

MARGIN = 1.0 # xi
WEIGHT = 0.0 # lambda

TRIALS = 1
CYCLES = 1

EPOCH = 50
LR = 0.1
MILESTONES = [25, 35]
EPOCHL = 40

MOMENTUM = 0.9
WDECAY = 5e-4
'''