def __init__(
self,
state_space_parameters,
epsilon,
data_path='./MNIST',
state=None,
qstore=None,
# replay_dictionary = pd.DataFrame(columns=['net',
# 'accuracy_best_val',
# 'accuracy_last_val', # uncomment while actual training
# 'epsilon',
# 'train_flag'])):
replay_dictionary=pd.DataFrame(columns=[
'net', 'loss_inverse', 'loss', 'epsilon', 'computeLoss_flag'
])):
self.state_list = []
self.data_path = data_path
# self.bucketed_state_list = []
self.state_space_parameters = state_space_parameters
self.enum = se.StateEnumerator(state_space_parameters)
self.stringutils = StateStringUtils(state_space_parameters)
self.state = se.State('start', 0, 1, 0, 0,
state_space_parameters.image_size, 0,
0) if not state else state
# self.bucketed_state = self.enum.bucket_state(self.state)
self.qstore = QValues() if not qstore else qstore
self.replay_dictionary = replay_dictionary
self.epsilon = epsilon
def __init__(
self,
state_space_parameters,
epsilon,
WeightInitializer=None,
device=None,
args=None,
save_path=None,
state=None,
qstore=None,
replaydict=None,
replay_dictionary=pd.DataFrame(columns=[
'net_disc', 'input_size', 'reward', 'epsilon', 'train_flag'
])):
self.state_list = []
self.state_space_parameters = state_space_parameters
self.args = args
self.enum = se.StateEnumerator(state_space_parameters, args)
self.stringutils = StateStringUtils(state_space_parameters, args)
self.state = se.State('start', 0, 1, 0, 0, args.patch_size, 0,
0) if not state else state
self.qstore = QValues() if not qstore else qstore
if type(qstore) is not type(None):
self.qstore.load_q_values(qstore_)
self.replay_dictionary = pd.read_csv(replaydict, index_col=0)
else:
self.replay_dictionary = replay_dictionary
self.epsilon = epsilon
self.WeightInitializer = WeightInitializer
self.device = device
self.gpu_mem_0 = GPUMem(torch.device('cuda') == self.device)
self.save_path = save_path
# TODO: hard-coded arc no. to resume from if epsilon < 1
self.count = args.continue_ite - 1
def __init__(
self,
premise,
hypothesis,
epsilon,
state_space_parameters,
state=None,
qstore=None,
replaydict=None,
WeightInitializer=None,
device=None,
replay_dictionary=pd.DataFrame(columns=[
'path', 'epsilon', 'accuracy_best_val', 'accuracy_last_val',
'accuracy_best_test', 'accuracy_last_test', 'ix_q_value_update'
])):
self.state_list = []
self.state_space_parameters = state_space_parameters
self.enumerator = se.StateEnumerator(state_space_parameters)
self.state = se.State('start', premise, 0, self.state_list)
self.qstore = QValues()
if type(qstore) is not type(None):
self.qstore.laod_q_values(qstore)
self.replay_dictionary = pd.read_csv(replaydict, index_col=0)
else:
self.replay_dictionary = replay_dictionary
self.epsilon = epsilon
self.WeightInitializer = WeightInitializer
self.device = device
def __init__(self,
state_space_parameters,
epsilon,
state=None,
qstore=None,
replay_dictionary=pd.DataFrame(columns=[
'net', 'accuracy_best_val', 'accuracy_last_val',
'accuracy_best_test', 'accuracy_last_test',
'ix_q_value_update', 'epsilon'
])):
self.state_list = []
self.state_space_parameters = state_space_parameters
# Class that will expand states for us
self.enum = se.StateEnumerator(state_space_parameters)
self.stringutils = StateStringUtils(state_space_parameters)
# Starting State
self.state = se.State('start', 0, 1, 0, 0,
state_space_parameters.image_size, 0,
0) if not state else state
self.bucketed_state = self.enum.bucket_state(self.state)
# Cached Q-Values -- used for q learning update and transition
self.qstore = QValues() if not qstore else qstore
self.replay_dictionary = replay_dictionary
self.epsilon = epsilon # epsilon: parameter for epsilon greedy strategy
def __init__(self, state_space_parameters, network_number, qstore=None):
self.state_list = []
self.state_space_parameters = state_space_parameters
# Class that will expand states for us
self.enum = se.StateEnumerator(state_space_parameters)
self.stringutils = StateStringUtils(state_space_parameters)
# Starting State
self.state = se.State('start', 0, 1, 0, 0, state_space_parameters.image_size, 0, 0, 0, 0, 0, 0, 0, 0, 0)# if not state else state
self.bucketed_state = self.enum.bucket_state(self.state)
# Cached Q-Values -- used for q learning update and transition
self.qstore = QValues() if not qstore else qstore
# self.replay_dictionary = replay_dictionary
# self.epsilon=epsilon # epsilon: parameter for epsilon greedy strategy
self.network_number = network_number
def __init__(self, state_space_parameters):
self.image_size = state_space_parameters.image_size
self.output_number = state_space_parameters.output_states
self.enum = se.StateEnumerator(state_space_parameters)
def __init__(self, state_space_parameters, args):
# TODO: decide on patch-size
self.image_size = args.patch_size
# TODO: unnecessary to instantiate StateEnumerator again
self.enum = se.StateEnumerator(state_space_parameters, args)
self.ssp = state_space_parameters
def __init__(self, state_space_parameters, args):
self.image_size = args.patch_size
self.ssp = state_space_parameters
self.enum = se.StateEnumerator(state_space_parameters, args)