def baselevel_activation(self, times, d, now=None):
''' Base level activation.
Input: Vector of calling times for a chunk.
Like [45023, 45046, 45102]
Formula to compute baselevel activation:
B_i = ln(sum_from_j=1_to_n(t_j^(-d))) with
n: The number of presentations for chunk i = length of times.
tj: The time since the jth presentation.
d: The decay parameter which is set using the
:bll (base-level learning) parameter.
This parameter is almost always set to 0.5.
Values range around 0
- activation falls
+ activation rises
'''
if not now:
now = utilities.milliseconds(datetime.datetime.now())
summed = 0
for j in range(len(times)):
diff = (now - times[j]) / 1000
summed += math.pow(diff, -d)
return math.log(summed)
def paired_associate(self):
''' Simulating the paired associate example.
'''
from environment import Word, Pair
pairs = [Pair('bank', '0'), Pair('card', '1'), Pair('dart', '2'),
Pair('face', '3'), Pair('game', '4'),
Pair('hand', '5'), Pair('jack', '6'), Pair('king', '7'),
Pair('lamb', '8'), Pair('mask', '9'),
Pair('neck', '0'), Pair('pipe', '1'), Pair('guip', '2'),
Pair('rope', '3'), Pair('sock', '4'),
Pair('tent', '5'), Pair('vent', '6'), Pair('wall', '7'),
Pair('xray', '8'), Pair('zinc', '9')]
nr_runs = 8
now = utilities.milliseconds(datetime.datetime.now())
total_recalled = [0 for i in range(nr_runs)]
total_probs = [0 for i in range(nr_runs)]
for k in range(100):
times = self.get_times(nr_runs, now)
probs = self.get_probabilities(times, now)
mean_probs = self.get_mean(probs)
recalled = self.simulate_run(probs)
for j in range(len(recalled)):
total_recalled[j] += recalled[j]
total_probs[j] += mean_probs[j]
for i in range(nr_runs):
print i+1, total_recalled[i] / 100, total_probs[i] / 100
def formulate_expectation(self, word):
now = utilities.milliseconds(datetime.datetime.now())
activation = self.activation(word.times)
retrieval_prob = self.retrieval_probability(activation, CogModule.THRESHOLD, CogModule.NOISE)
retrieval_latency = self.retrieval_latency(CogModule.LATENCY, activation)
self.expectation = self.get_expectation_name(retrieval_prob)
status = str(retrieval_prob) + ': '
emotion = None
if self.expectation == 'positive':
status += 'Expecting right answer.'
emotion = CogModule.EXPECT_POS[0]
elif self.expectation == 'none':
status += 'Expecting nothing.'
elif self.expectation == 'negative':
status += 'Expecting wrong answer'
emotion = CogModule.EXPECT_NEG[0]
else:
print 'Wrong expectation value', self.expectation
self.logger.log(' Formulate expectation: prob={0:.2f}% latency={1:.2f}s : ({2},{3})'.format(retrieval_prob, retrieval_latency, self.expectation, emotion))
return (status, utilities.emotion_by_name(emotion))
def optimized_learning(self, times, d, now=None):
''' Formula for base level learning. Optimizes computation time.
B_i = ln(n / (1-d)) - d * ln(L) with
n = nr of presentations of chunk i
L = The lifetime of chunk i (the time since its creation)
d = The decay parameter
'''
if not now:
now = utilities.milliseconds(datetime.datetime.now())
n = len(times)
L = now - times[0]
return math.log(n / (1 - d)) - d * math.log(L)
def present_number(self, now):
'''
'''
time_delta = now-self.start_time
print time_delta, ': PRESENT NR', time_delta.seconds
# Check if answer has been given. If not evaluate.
if not self.answer_given:
self.logger.log(' No answer given')
self.evaluate('-1', now)
if 0 <= self.index and self.index <= len(self.pairs):
number = self.pairs[self.index].number
self.pairs[self.index].number_called(utilities.milliseconds(now))
self.index += 1
return self.agent.present_number(number)
else:
print 'Index Error'
def activation(self, times, now=None):
''' Computes the activation for the given chunk.
'''
if not times:
return 0.0
d = CogModule.DECAY_RATE
s = CogModule.NOISE
if not now:
now = utilities.milliseconds(datetime.datetime.now())
if CogModule.FUNCTION == 'optimized':
base_activation = self.optimized_learning(times, d)
else:
base_activation = self.baselevel_activation(times, d)
noise = self.logistic_distribution(s)
return base_activation + noise
def present_word(self, now):
'''
'''
time_delta = now-self.start_time
print time_delta, ': PRESENT WORD', time_delta.seconds
if self.has_next():
word = self.pairs[self.index].word
number = self.pairs[self.index].number
self.logger.log('\nTask [{0} : {1}] @ {2}s'.format(
word.word, number.word, time_delta))
self.pairs[self.index].word_called(utilities.milliseconds(now))
self.start_time_answer = now
self.answer_given = False
return self.agent.present_word(word, number)
else:
print 'Index Error'
