def get_scalability(num_s):
process = psutil.Process(os.getpid())
scores = []
# Setup Block
encoders = []
pattern_poolers = []
sequence_learners = []
encoder = ScalarEncoder(min_val=0.0, max_val=1.0, num_s=1024, num_as=128)
pattern_pooler = PatternPooler(num_s=num_s,
num_as=8,
pct_pool=0.8,
pct_conn=0.8,
pct_learn=0.25)
sequence_learner = SequenceLearner(num_spc=10,
num_dps=10,
num_rpd=12,
d_thresh=6)
pattern_pooler.input.add_child(encoder.output)
sequence_learner.input.add_child(pattern_pooler.output)
num_coincidence_detectors = num_s + (num_s * 10 * 10
) # pattern_pooler + sequence_learner
# Get initialization time and memory usage
t0 = time.time()
encoder.compute(data[0])
pattern_pooler.compute(learn=True)
sequence_learner.compute(learn=True)
score = sequence_learner.get_score()
scores.append(score)
t1 = time.time()
init_time = t1 - t0
num_bytes = process.memory_info().rss
# Get compute time
t0 = time.time()
for i in range(1, len(data)):
encoder.compute(data[i])
pattern_pooler.compute(learn=True)
sequence_learner.compute(learn=True)
score = sequence_learner.get_score()
scores.append(score)
t1 = time.time()
comp_time = t1 - t0
# Test Results
np.testing.assert_array_equal(np.array(scores), expected_scores)
return [num_coincidence_detectors, num_bytes, init_time, comp_time]
def test_sequence_learner():
e = ScalarEncoder(min_val=0.0, max_val=1.0, num_s=64, num_as=8)
sl = SequenceLearner(num_spc=10,
num_dps=10,
num_rpd=12,
d_thresh=6,
perm_thr=1,
perm_inc=1,
perm_dec=0)
sl.input.add_child(e.output)
data = [
0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0,
0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0,
1.0, 1.0
]
expect_scores = np.array([
1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0
])
actual_scores = np.array([0.0 for i in range(len(data))])
for i in range(len(data)):
e.compute(data[i])
sl.compute(True)
actual_scores[i] = sl.get_score()
np.testing.assert_array_equal(actual_scores, expect_scores)
def test_sequence_learner_sine():
e = ScalarEncoder(
min_val=0.0, # minimum input value
max_val=1.0, # maximum input value
num_s=64, # number of statelets
num_as=8) # number of active statelets
sl = SequenceLearner(
num_spc=10, # number of statelets per column
num_dps=10, # number of coincidence detectors per statelet
num_rpd=24, # number of receptors per coincidence detector
d_thresh=6, # coincidence detector threshold
perm_thr=1, # receptor permanence threshold
perm_inc=1, # receptor permanence increment
perm_dec=0) # receptor permanence decrement
sl.input.add_child(e.output)
values = [
0.50, 0.79, 0.98, 0.98, 0.79, 0.50, 0.21, 0.02, 0.02, 0.21, 0.50, 0.79,
0.98, 0.98, 0.79, 0.50, 0.21, 0.02, 0.02, 0.21, 0.50, 0.79, 0.98, 0.98,
0.79, 0.50, 0.21, 0.02, 0.02, 0.21
]
actual_scores = np.array([0.0 for i in range(len(values))])
expect_scores = np.array([
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0
])
for i in range(len(values)):
e.compute(value=values[i])
sl.compute(learn=True)
actual_scores[i] = sl.get_score()
np.testing.assert_array_equal(actual_scores, expect_scores)
pct_conn=0.8, # initially connected percentage
pct_learn=0.25) # learn percentage 0.25
sl.input.add_child(e.output)
pc.input.add_child(sl.output)
aed = AbnormalEventDetector(5, 5)
print('val scr lbl prob ae output_active_statelets')
for i in range(len(values)):
e.compute(value=values[i])
sl.compute(learn=True)
pc.compute(learn=False)
score = sl.get_score()
probs = pc.get_probabilities()
abnormal_event = aed.compute(score)
if abnormal_event:
for _ in range(50):
pc.compute(label=new_label, learn=True)
new_label += 1
winner = np.argmax(probs)
winner_str = '-'
#winner_str = str(winner)
if probs[winner] >= 0.75:
winner_str = str(winner)
pp_0.input.add_child(se_0.output)
pp_1.input.add_child(se_1.output)
pp_2.input.add_child(pp_0.output)
pp_2.input.add_child(pp_1.output)
sl_0.input.add_child(pp_0.output)
sl_1.input.add_child(pp_1.output)
sl_2.input.add_child(pp_2.output)
# loop through data
for i in range(num_inputs):
# compute
se_0.compute(inputs_0[i])
se_1.compute(inputs_1[i])
pp_0.compute()
pp_1.compute()
pp_2.compute()
sl_0.compute()
sl_1.compute()
sl_2.compute()
# get scores
scores_0[i] = sl_0.get_score()
scores_1[i] = sl_1.get_score()
scores_2[i] = sl_2.get_score()
# print output
print("in0, in1, sc0, sc1, sc2")
for i in range(num_inputs):
print("%0.1f, %0.1f, %0.1f, %0.1f, %0.1f" % (inputs_0[i], inputs_1[i], scores_0[i], scores_1[i], scores_2[i]))
def multiple_prior_contexts():
experiment_name = 'multiple_prior_contexts'
directory = './' + experiment_name
mkdir_p(directory)
print()
print('experiment=\'%s\'' % (experiment_name))
e = SymbolsEncoder(
max_symbols=26, # maximum number of symbols
num_s=208) # number of statelets
sl = SequenceLearner(
num_spc=10, # number of statelets per column
num_dps=50, # number of coincidence detectors per statelet
num_rpd=12, # number of receptors per coincidence detector
d_thresh=6, # coincidence detector threshold
perm_thr=1, # receptor permanence threshold
perm_inc=1, # receptor permanence increment
perm_dec=0) # receptor permanence decrement
sl.input.add_child(e.output)
values = [
'a', 'z', 'a', 'z', 'a', 'z', 'b', 'z', 'b', 'z', 'b', 'z', 'c', 'z',
'c', 'z', 'c', 'z', 'd', 'z', 'd', 'z', 'd', 'z', 'e', 'z', 'e', 'z',
'e', 'z', 'f', 'z', 'f', 'z', 'f', 'z', 'g', 'z', 'g', 'z', 'g', 'z',
'h', 'z', 'h', 'z', 'h', 'z', 'i', 'z', 'i', 'z', 'i', 'z', 'j', 'z',
'j', 'z', 'j', 'z', 'k', 'z', 'k', 'z', 'k', 'z', 'l', 'z', 'l', 'z',
'l', 'z', 'm', 'z', 'm', 'z', 'm', 'z', 'n', 'z', 'n', 'z', 'n', 'z',
'o', 'z', 'o', 'z', 'o', 'z', 'p', 'z', 'p', 'z', 'p', 'z', 'q', 'z',
'q', 'z', 'q', 'z', 'r', 'z', 'r', 'z', 'r', 'z', 's', 'z', 's', 'z',
's', 'z', 't', 'z', 't', 'z', 't', 'z', 'u', 'z', 'u', 'z', 'u', 'z',
'v', 'z', 'v', 'z', 'v', 'z', 'w', 'z', 'w', 'z', 'w', 'z', 'x', 'z',
'x', 'z', 'x', 'z', 'y', 'z', 'y', 'z', 'y', 'z'
]
le = preprocessing.LabelEncoder()
le.fit(values)
int_values = le.transform(values)
scores = [0.0 for _ in range(len(values))]
count_s_acts = [0 for _ in range(len(values))]
count_s_hist = [0 for _ in range(len(values))]
count_cs = [0 for _ in range(len(values))]
hidden_s_usage = [0 for _ in range(2240)]
output_s_usage = [0 for _ in range(2240)]
print('val scr s_act s_his cs active output statelets')
for i in range(len(int_values)):
e.compute(value=int_values[i])
sl.compute(learn=True)
# update information
hidden_s_bits = sl.hidden.bits
hidden_s_acts = sl.hidden.acts
output_s_bits = sl.output.bits
output_s_acts = sl.output.acts
scores[i] = sl.get_score()
count_s_acts[i] = len(output_s_acts)
count_s_hist[i] = sl.get_historical_count()
count_cs[i] = sl.get_coincidence_set_count()
# update statelet usage
for s in range(len(output_s_usage)):
hidden_s_usage[s] += hidden_s_bits[s]
output_s_usage[s] += output_s_bits[s]
# plot statelets
if (i + 1) % 6 == 0:
title = values[i] + '_' + values[i - 1]
plot_statelets(directory, 'hidden_' + title, hidden_s_bits)
plot_statelets(directory, 'output_' + title, output_s_bits)
# print information
output_s_acts_str = '[' + ', '.join(
str(act).rjust(4) for act in output_s_acts) + ']'
print('{0:>3} {1:0.1f} {2:5d} {3:5d} {4:4d} {5:>4}'.format(
values[i], scores[i], count_s_acts[i], count_s_hist[i],
count_cs[i], output_s_acts_str))
# plot information
plot_results(directory, 'results', values, scores, count_s_acts,
count_s_hist, count_cs, 600, 2200)
plot_statelet_usage(directory, 'hidden', hidden_s_usage, 75)
plot_statelet_usage(directory, 'output', output_s_usage, 75)
'''
class AbnormalityDetector():
def __init__(
self,
configs=(), # block configuration
min_val=-1.0, # minumum value
max_val=1.0, # maximum value
num_i=1024, # ScalarEncoder number of statelets
num_ai=128, # ScalarEncoder number of active statelets
num_s=512, # PatternPooler number of statelets
num_as=8, # PatternPooler number of active statelets
num_spc=10, # SequenceLearner number of statelets per column
num_dps=10, # SequenceLearner number of coincidence detectors per statelet
num_rpd=12, # SequenceLearner number of receptors per coincidence detector
d_thresh=6, # SequenceLearner coincidence detector threshold
pct_pool=0.8, # PatternPooler pool percentage
pct_conn=0.5, # PatternPooler initial connection percentage
pct_learn=0.25): # PatternPooler learn percentage
self.min_val = min_val
self.max_val = max_val
# seed the random number generator
bb.seed(0)
# build blocks from config descriptions if given
blocks = get_blocks(configs)
self.encoders = blocks["encoders"]
self.pp = blocks["pattern_pooler"]
self.sl = blocks["sequence_learner"]
if len(self.encoders) == 0:
self.encoders.append(ScalarEncoder(min_val, max_val, num_i, num_ai))
if self.pp == None:
self.pp = PatternPooler(num_s, num_as, 20, 2, 1, pct_pool, pct_conn, pct_learn)
if self.sl == None:
self.sl = SequenceLearner(num_spc, num_dps, num_rpd, d_thresh, 1, 1, 0)
for encoder in self.encoders:
self.pp.input.add_child(encoder.output)
self.sl.input.add_child(self.pp.output)
self.initialized = False
def print_parameters(self):
for encoder in self.encoders:
encoder.print_parameters()
self.pp.print_parameters()
self.sl.print_parameters()
def save_memories(self, path='./', name='detector'):
self.pp.save_memories(path + name + "_pp.bin")
self.sl.save_memories(path + name + "_sl.bin")
def load_memories(self, path='./', name='detector'):
self.pp.load_memories(path + name + "_pp.bin")
self.sl.load_memories(path + name + "_sl.bin")
def compute(self, vectors=(), learn=True):
anoms = []
num_steps = 0xFFFFFFFF
num_measurands = len(vectors)
num_encoders = len(self.encoders)
if num_measurands != num_encoders:
print("Warning: compute() num_measurands != num_encoders")
return anoms
for vector in vectors:
len_vector = len(vector)
if len_vector < num_steps:
num_steps = len_vector
for e in range(num_encoders):
if isinstance(self.encoders[e], PersistenceEncoder):
self.encoders[e].reset()
limit_flag = 0
for s in range(num_steps):
for e in range(num_encoders):
value = vectors[e][s]
if value < self.min_val or value > self.max_val:
limit_flag = 1
self.encoders[e].compute(value)
self.pp.compute(learn)
self.sl.compute(learn)
if limit_flag == 1:
anoms.append(1.0)
else:
anoms.append(self.sl.get_score())
self.initialized = True
return anoms
def three_events(statelet_snapshots_on=False):
experiment_name = 'three_events'
directory = './' + experiment_name
mkdir_p(directory)
print()
print('experiment=\'%s\'' % (experiment_name))
NUM_S = 208
NUM_SPC = 10
TOTAL_NUM_S = NUM_S * NUM_SPC
e = SymbolsEncoder(
max_symbols=26, # maximum number of symbols
num_s=NUM_S) # number of statelets
sl = SequenceLearner(
num_spc=NUM_SPC, # number of statelets per column
num_dps=50, # number of coincidence detectors per statelet
num_rpd=12, # number of receptors per coincidence detector
d_thresh=6, # coincidence detector threshold
perm_thr=1, # receptor permanence threshold
perm_inc=1, # receptor permanence increment
perm_dec=0) # receptor permanence decrement
sl.input.add_child(e.output)
values = [
'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'b', 'c', 'd', 'e',
'f', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'f', 'e', 'd',
'c', 'b', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'b', 'c',
'd', 'c', 'b', 'a', 'a', 'a', 'a', 'a'
]
le = preprocessing.LabelEncoder()
le.fit(values)
int_values = le.transform(values)
scores = [0.0 for _ in range(len(values))]
count_s_output_acts = [0 for _ in range(len(values))]
count_s_hidden_acts = [0 for _ in range(len(values))]
count_s_hist = [0 for _ in range(len(values))]
count_cs = [0 for _ in range(len(values))]
hidden_s_usage = [0 for _ in range(TOTAL_NUM_S)]
output_s_usage = [0 for _ in range(TOTAL_NUM_S)]
print('val scr s_act s_his cs active output statelets')
for i in range(len(int_values)):
e.compute(value=int_values[i])
sl.compute(learn=True)
# update information
hidden_s_bits = sl.hidden.bits
hidden_s_acts = sl.hidden.acts
output_s_bits = sl.output.bits
output_s_acts = sl.output.acts
scores[i] = sl.get_score()
count_s_output_acts[i] = len(output_s_acts)
count_s_hidden_acts[i] = len(hidden_s_acts)
count_s_hist[i] = sl.get_historical_count()
count_cs[i] = sl.get_coincidence_set_count()
# update statelet usage
for s in range(len(output_s_usage)):
hidden_s_usage[s] += hidden_s_bits[s]
output_s_usage[s] += output_s_bits[s]
# plot statelets
if statelet_snapshots_on and (i + 1) % 5 == 0 or i == 43:
title = 'step_' + str(i) + '_' + values[i] + '_' + values[i - 1]
plot_statelets(directory, 'hidden_' + title, hidden_s_bits)
plot_statelets(directory, 'output_' + title, output_s_bits)
# print information
output_s_acts_str = '[' + ', '.join(
str(act).rjust(4) for act in output_s_acts) + ']'
print('{0:>3} {1:0.1f} {2:5d} {3:5d} {4:4d} {5:>4}'.format(
values[i], scores[i], count_s_output_acts[i], count_s_hist[i],
count_cs[i], output_s_acts_str))
# plot information
plot_results(directory, 'results', values, scores, count_s_output_acts,
count_s_hidden_acts, count_s_hist, count_cs, 400, 400)
plot_statelet_usage(directory, 'hidden', hidden_s_usage, 75)
plot_statelet_usage(directory, 'output', output_s_usage, 75)