def __init__(self, rnn_type, consolidate_moment, consolidate_threshold, nin, nout, ninp, nhid, nlayers,
max_memory_size, max_ltm_size, init_stm_size, lr, batch_size, clip, optimizer, ltm_reinstatement, stm_consolidation,
train_weights_before_predict, weights_trainer, residual_weights_trainer=None,
tie_weights=False, w_window=20, recent_window_size=5, dropout=0.2, is_cuda = True):
self.optimizers = []
self.consolidate_moment = consolidate_moment
self.consolidation_threshold = consolidate_threshold
self.reinstatement_threshold = consolidate_threshold
self.consolidate_moment_ticks = 1
self.residual_weights_trainer = None
self.ltm_size = 0
self.memory_size = 0
self.max_ltm_size = max_ltm_size
self.init_stm_size = init_stm_size
self.max_memory_size = max_memory_size
self.stm_consolidation = stm_consolidation
self.ltm_reinstatement = ltm_reinstatement
self.recent_window_size = recent_window_size
self.weight_history = []
self.weights_copied = Observable()
self.weights_moved = Observable()
self.ltm_size_updated = Observable()
super(CloneLearner, self).__init__(rnn_type, nin, nout, ninp, nhid, nlayers,
max_memory_size, lr, batch_size, clip, optimizer,
train_weights_before_predict, weights_trainer, tie_weights, w_window, dropout, is_cuda)
self.weights_updated.register(self.save_weights_history)
def __init__(self, rnn_type, nin, nout, ninp, nhid, nlayers,
max_memory_size, lr, batch_size, clip, optimizer,
train_weights_before_predict, weights_trainer,
learn_iterations, tie_weights, w_window, dropout, is_cuda):
criterion = nn.CrossEntropyLoss()
super(BaseMixtureOfRNNsLearner, self).__init__(criterion, nout,
learn_iterations)
self.optimizer_algorithm = optimizer
#general parameters
self.rnn_type = rnn_type
self.nin = nin
self.nout = nout
self.ninp = ninp
self.nhid = nhid
self.max_memory_size = max_memory_size
self.dropout = dropout
self.rnns = []
self.ids = []
self.id_count = 0
self.lr = lr
self.batch_size = batch_size
self.hidden = []
self.clip = clip
self.window_size = w_window
self.nlayers = nlayers
self.train_weights_before_predict = train_weights_before_predict
self.cuda_available = is_cuda
self.outputs = torch.zeros(self.max_memory_size, 20, self.batch_size,
self.nout)
self.weights_trainer = weights_trainer
self.tie_weights = tie_weights
self.weights_added = Observable()
self.weights_removed = Observable()
self.weights_updated = Observable()
self._initialize_model()
def test_many_events_with_dictionary():
subject = Observable()
subject.on({ # setting events of an observable with dictionary
'one': one_handler,
'two': two_handler,
'custom': CustomEvent(),
'many': [one_handler, two_handler]})
subject.trigger('one two custom many', 1, 2, a=3, b=4)
def test_custom_event():
subject = Observable()
custom_event = CustomEvent()
subject.on('custom', custom_event)
subject.on(['custom', 'custom_alias'], custom_event) # same reference
assert(subject.custom and subject.custom_alias is custom_event)
assert(one_handler and two_handler in subject.custom.handlers)
# update an event
subject.on('custom', CustomEvent())
assert(subject.custom is not custom_event)
def test_event_with_many_handlers_attached():
subject = Observable()
subject.on('many', [one_handler, two_handler])
assert(subject.events['many'] is subject.many)
assert(one_handler and two_handler in subject.many.handlers)
# remove a handler
subject.off('many', one_handler)
assert(one_handler not in subject.many.handlers)
# remove the event
subject.off('many')
assert('many' not in subject.events)
with raises(AttributeError):
subject.many
def test_event_with_handler_attached():
subject = Observable()
subject.on('one', one_handler)
subject.on('two', two_handler)
assert(subject.events['one'] is subject.one)
assert(subject.events['two'] is subject.two)
# trigger
subject.one(1, 2, a=3, b=4)
subject.one.trigger(1, 2, a=3, b=4)
subject.trigger('one', 1, 2, a=3, b=4)
subject.events['one'].trigger(1, 2, a=3, b=4)
subject.trigger('one two', 1, 2, a=3, b=4)
subject.trigger(['one', 'two'], 1, 2, a=3, b=4)
def test_update(self):
observable = Observable()
observer = Observer(1)
observable.addObserver(observer)
observable.updateObservers(2)
self.assertEqual(observer.data, 2)
"""Alarm Event names"""
from observer import Observable
import logging
LOGGER = logging.getLogger('alarm')
# Events
MOTION_DETECTED = 'motion_detected'
# Basic handlers
def log_it(*args, **kwargs):
LOGGER.debug("%s, %s", args, kwargs)
# Setup events
observable = Observable()
observable.on(MOTION_DETECTED, [log_it])
def test_no_callable_handler():
subject = Observable()
with raises(TypeError):
subject.on('error', NotCallable())
def test_events_with_same_handlers_attached():
subject = Observable()
subject.on('many many2', [one_handler, two_handler])
assert(subject.many is not subject.many2)
assert(one_handler and two_handler in subject.many.handlers)
assert(one_handler and two_handler in subject.many2.handlers)
def test_events_with_same_handler_attached():
subject = Observable()
subject.on(['one', 'one2'], one_handler)
assert(subject.one is not subject.one2)
assert(one_handler in subject.one.handlers)
assert(one_handler in subject.one2.handlers)
###############################################################################
# Build the model
###############################################################################
vocsize = len(corpora.vocabulary)
print('Vocabulary size:', vocsize)
print('Creating learner...')
learner = get_learner(args, vocsize)
print(f'Number of parameters: {learner.get_num_parameters()/1e6:1.2f}M')
if args.cuda:
learner.cuda()
if args.load_from:
learner.load_from(torch.load(args.load_from))
domain_switched = Observable()
timestep_updated = Observable()
learner.train_mode()
if args.log_weights:
weight_logger = log_utils.WeightsLogger(learner, args.log_weights)
domain_switched.register(weight_logger.domain_switched)
timestep_updated.register(weight_logger.timestep_updated)
###############################################################################
# Training code
###############################################################################
def generate_text(sequence_id, position, sequence_type, sequence_type_name):
learner.evaluate_mode()
input = Variable(torch.rand(1, 1).mul(len(corpora.vocabulary)).long(),
self.plotly_user_config['plotly_streaming_tokens'][0])
self.stream.open()
def notify(self, observable, *args, **kwargs):
# write to plotly
if 'value' in kwargs:
# self.stream.write({'x': datetime.datetime.now(), 'y': kwargs['value']})
self.stream.write({'x': kwargs['timestamp'], 'y': kwargs['value']})
# import plotly.plotly as py
# from plotly.graph_objs import *
# trace0 = Scatter(
# x=[1, 2, 3, 4],
# y=[10, 15, 13, 17]
# )
# trace1 = Scatter(
# x=[1, 2, 3, 4],
# y=[16, 5, 11, 9]
# )
# data = Data([trace0, trace1])
# unique_url = py.plot(data, filename = 'basic-line')
if __name__ == '__main__':
subject = Observable()
plyobs = plotlyObserver(subject, 'config.json', 'test stream plot')
subject.notify_observers(value=45)
subject.notify_observers(value=55)
