def main():
x = 10
y = 10
steps = 10000
world = np.array([i for i in xrange(625)])
world.resize((25, 25))
spInputSize = 21*21
sp = SpatialPooler(
inputDimensions=(spInputSize,),
columnDimensions=(25,),
potentialRadius=spInputSize,
numActiveColumnsPerInhArea=1,
synPermActiveInc=0.1,
synPermInactiveDec=0.5,
boostStrength=1.0,
)
csFields = generateCenterSurroundFields()
output = np.zeros((25,), dtype=np.uint32)
for _ in xrange(steps):
active = getActive(world, x, y)
assert len(active) == 25, "{}, {}: {}".format(x, y, active)
activeInput = np.zeros((625,), dtype=np.uint32)
for v in active:
activeInput[v] = 1
centerSurround = processCenterSurround(csFields, activeInput)
print centerSurround
sp.compute(centerSurround, True, output)
x, y = getNewLocation(x, y, 25, 2, False)
for i in xrange(25):
permanence = np.zeros((spInputSize,))
sp.getPermanence(i, permanence)
plt.imshow(permanence.reshape((21, 21)), cmap="hot", interpolation="nearest")
plt.show()
def test_whether_is_the_same_as_spatial_pooler(self):
"""
Naive reality check for the encoding function
of the lateral pooler implementation.
"""
n = 1024
m = 784
d = 100
w = 20
X = np.random.randint(0,2,size=(m,d))
Y_nup = np.zeros((n,d))
Y_lat = np.zeros((n,d))
params_nup = {
"inputDimensions": [m,1],
"columnDimensions": [n,1],
"potentialRadius": n,
"potentialPct": 1.0,
"globalInhibition": True,
"localAreaDensity": -1.0,
"numActiveColumnsPerInhArea": w,
"stimulusThreshold": 0,
"synPermInactiveDec": 0.05,
"synPermActiveInc" : 0.1,
"synPermConnected" : 0.5,
"minPctOverlapDutyCycle": 0.001,
"dutyCyclePeriod": 1000,
"boostStrength" : 100.0,
"seed": 1936 }
params_lat = params_nup.copy()
params_lat["lateralLearningRate"] = 0.0
params_lat["enforceDesiredWeight"] = False
sp_nup = SpatialPooler(**params_nup)
sp_lat = LateralPooler(**params_lat)
for t in range(d):
sp_nup.compute(X[:,t], False, Y_nup[:,t])
sp_lat.compute(X[:,t], False, Y_lat[:,t])
self.assertTrue(np.all(Y_nup == Y_lat),
"Produces wrong output even without learning.")
for t in range(d):
sp_nup.compute(X[:,t], True, Y_nup[:,t])
sp_lat.compute(X[:,t], True, Y_lat[:,t])
self.assertTrue(np.all(Y_nup == Y_lat),
"Wrong outputs, something diverges during learning.")
W_nup = get_W(sp_nup)
W_lat = get_W(sp_lat)
self.assertTrue(np.all(W_nup == W_lat),
"Wrong synaptic weights, something diverges during learning.")
def create_sp(self):
sp = SpatialPooler(
# How large the input encoding will be.
inputDimensions=(len(self.encoding[0])),
# How many mini-columns will be in the Spatial Pooler.
columnDimensions=(2048),
# What percent of the columns's receptive field is
# available for potential synapses?
potentialPct=0.85,
# This means that the input space has no topology.
globalInhibition=True,
localAreaDensity=-1.0,
# Roughly 2%, giving that there is only one inhibition area because
# we have turned on globalInhibition (40 / 2048 = 0.0195)
numActiveColumnsPerInhArea=40.0,
# How quickly synapses grow and degrade.
synPermInactiveDec=0.005,
synPermActiveInc=0.04,
synPermConnected=0.1,
# boostStrength controls the strength of boosting. Boosting
# encourages efficient usage of SP columns.
boostStrength=3.0,
# Random number generator seed.
seed=1956,
# Determines if inputs at the beginning and end of an input dimension
# should be considered neighbors when mapping columns to inputs.
wrapAround=False)
return sp
def init_imgs(write_output):
#problems = parse_images.get_problems(folder_name)
folder_name = 'Data/raw'
problems = []
for sub_folder in os.listdir(folder_name):
f = os.path.join(folder_name, sub_folder)
problems += parse_images.get_problems(f)
print(len(problems))
for problem in problems:
problem['Input'] = problem['Input'].reshape((3, -1))
problem['Output'] = problem['Output'].reshape((6, -1))
# dimenstions
num_cols1 = len(problems[0]['Input'][0])
tm_cols = num_cols1
layers = []
if not write_output:
num_cols2 = num_cols1 / 2
num_cols3 = num_cols2 / 2
sp1 = SpatialPooler(inputDimensions=(num_cols1, ),
columnDimensions=(num_cols2, ),
numActiveColumnsPerInhArea=-1,
localAreaDensity=0.05)
sp2 = SpatialPooler(inputDimensions=(num_cols2, ),
columnDimensions=(num_cols3, ),
numActiveColumnsPerInhArea=-1,
localAreaDensity=0.05)
tm_cols = num_cols3
layers = [(sp1, sp_compute), (sp2, sp_compute)]
bckTM = BTM(numberOfCols=tm_cols,
cellsPerColumn=10,
initialPerm=0.5,
connectedPerm=0.5,
minThreshold=10,
newSynapseCount=10,
activationThreshold=10,
pamLength=10)
layers += [(bckTM, bk_tm_compute)]
return (layers, problems)
def testCompatibilityCppPyDirectCall1D(self):
"""Check SP implementations have same behavior with 1D input."""
pySp = PySpatialPooler(inputDimensions=[121], columnDimensions=[300])
cppSp = CPPSpatialPooler(inputDimensions=[121], columnDimensions=[300])
data = numpy.zeros([121], dtype=uintType)
for i in range(21):
data[i] = 1
nCols = 300
d1 = numpy.zeros(nCols, dtype=uintType)
d2 = numpy.zeros(nCols, dtype=uintType)
pySp.compute(data, True, d1) # learn
cppSp.compute(data, True, d2)
d1 = d1.nonzero()[0].tolist()
d2 = d2.nonzero()[0].tolist()
self.assertListEqual(
d1, d2, "SP outputs are not equal: \n%s \n%s" % (str(d1), str(d2)))
def main():
x = 10
y = 10
steps = 10000
history = []
world = np.array([i for i in xrange(625)])
world.resize((25, 25))
sp = SpatialPooler(
inputDimensions=(625, ),
columnDimensions=(25, ),
potentialRadius=625,
numActiveColumnsPerInhArea=1,
)
output = np.zeros((25, ), dtype=np.uint32)
for _ in xrange(steps):
active = getActive(world, x, y)
assert len(active) == 25, "{}, {}: {}".format(x, y, active)
activeInput = np.zeros((625, ), dtype=np.uint32)
for v in active:
activeInput[v] = 1
history.append(active)
sp.compute(activeInput, True, output)
x, y = getNewLocation(x, y, 25, 2, False)
for i in xrange(25):
permanence = np.zeros((625, ))
sp.getPermanence(i, permanence)
plt.imshow(permanence.reshape((25, 25)),
cmap="hot",
interpolation="nearest")
plt.show()
def runTrial(ww, numColumns, potentialPct, inc, dec, mpo, dutyCycle, boost, steps, rr, spW, stimulusThreshold, connected, stepSize, jumpProb, directionStability):
ws = ww ** 2
x = 10
y = 10
locationHeatmap = np.zeros((ww, ww))
history = []
world = np.array([i for i in xrange(ws)])
world.resize((ww, ww))
sp = SpatialPooler(
inputDimensions=(ws,),
columnDimensions=(numColumns,),
potentialRadius=ws,
potentialPct=potentialPct,
numActiveColumnsPerInhArea=spW,
stimulusThreshold=stimulusThreshold,
synPermActiveInc=inc,
synPermInactiveDec=dec,
synPermConnected=connected,
minPctOverlapDutyCycle=mpo,
dutyCyclePeriod=dutyCycle,
boostStrength=boost,
seed=1936,
globalInhibition=True,
)
output = np.zeros((numColumns,), dtype=np.uint32)
direction = 0
for i in xrange(steps):
locationHeatmap[x][y] += 1
active = getActive(world, ww, x, y, rr)
history.append(active)
activeInput = np.zeros((ws,), dtype=np.uint32)
for v in active:
activeInput[v] = 1
sp.compute(activeInput, True, output)
x, y, direction = getNewLocation(x, y, ww, rr, wrap=True, locationHeatmap=locationHeatmap, stepSize=stepSize, jumpProb=jumpProb, direction=direction, directionStability=directionStability)
if (i + 1) % 100 == 0:
saveImage(history, ws, ww, numColumns, locationHeatmap, potentialPct, inc, dec, mpo, dutyCycle, boost, rr, spW, i+1, sp)
saveImage(history, ws, ww, numColumns, locationHeatmap, potentialPct, inc, dec, mpo, dutyCycle, boost, rr, spW, steps, sp)
def definir_SP(SIZE_ENCODER_):
"""
retorna a classe sp
"""
N_COLUMNS= 2048
sp = SpatialPooler(
inputDimensions = (SIZE_ENCODER_,),
columnDimensions = ( N_COLUMNS,), # in this case we will use 2048 mini-columns distributed in a "linear array" ...
potentialRadius = SIZE_ENCODER_, # i set the potential radius of each mini-column as the whole ...
#input space
potentialPct = 0.8, # how many bits on the input space that should have some permanence with each mini-column
## attention: having a permanence value doesn't mean that it will be connected, to be connected the "connection...
## force" / permanence needs to be higher than the threshold.
globalInhibition = True, # means that the winning columns are selected in the neighborhood, though in this code...
#we're dealing as if all the columns are neigbhors with one another
localAreaDensity = -1.0,
numActiveColumnsPerInhArea = NUM_ACTIVE_COLUMNS,
stimulusThreshold = 0,
##Well, if we set the number of active columns per input, than there is no need to set an stimulusThreshold
##First = because the simulusTHreshold will be already set as the sum of permanences of the 40th column...@@NEED TO CHECK
##any other mini-column with less than it won't be active on this input
synPermInactiveDec = 0.0005, #if a column is active, the off (that aren't 1 in input) bits which it is connected ...
# will have a decrement on the "synapse force"/permance @@
synPermActiveInc = 0.003,#if a column is active, the on (that are 1 in input) bits which it is connected ...
# will have a increment on the "synapse force"/permance @@
synPermConnected = 0.2, #how much the "strength" of the connection between the on bit and the mini-column ...
#needs to be for they to be connected. @@
# @@ what needs to be checked is if the bits with synPermConnected < 0.1 will decrement ou increment when a column ...
# is active - but i think they will increment
minPctOverlapDutyCycle = 0.001, # number between 0 and 1.0, used to set a floor on how often a column
#...should have at least stimulusThreshold active inputs
dutyCyclePeriod = 100, # how many "inputs seen" this should happen
boostStrength = 0.01,
seed = 47,
spVerbosity = 0,
wrapAround = False
)
return sp
def test_whether_is_the_same_as_spatial_pooler(self):
"""
Naive reality check for the encoding function
of the lateral pooler implementation.
"""
n = 1024
m = 784
d = 100
w = 20
X = np.random.randint(0, 2, size=(m, d))
Y_nup = np.zeros((n, d))
Y_lat = np.zeros((n, d))
params_nup = {
"inputDimensions": [m, 1],
"columnDimensions": [n, 1],
"potentialRadius": n,
"potentialPct": 1.0,
"globalInhibition": True,
"localAreaDensity": -1.0,
"numActiveColumnsPerInhArea": w,
"stimulusThreshold": 0,
"synPermInactiveDec": 0.05,
"synPermActiveInc": 0.1,
"synPermConnected": 0.5,
"minPctOverlapDutyCycle": 0.001,
"dutyCyclePeriod": 1000,
"boostStrength": 100.0,
"seed": 1936
}
params_lat = params_nup.copy()
params_lat["lateralLearningRate"] = 0.0
params_lat["enforceDesiredWeight"] = False
sp_nup = SpatialPooler(**params_nup)
sp_lat = LateralPooler(**params_lat)
for t in range(d):
sp_nup.compute(X[:, t], False, Y_nup[:, t])
sp_lat.compute(X[:, t], False, Y_lat[:, t])
self.assertTrue(np.all(Y_nup == Y_lat),
"Produces wrong output even without learning.")
for t in range(d):
sp_nup.compute(X[:, t], True, Y_nup[:, t])
sp_lat.compute(X[:, t], True, Y_lat[:, t])
self.assertTrue(np.all(Y_nup == Y_lat),
"Wrong outputs, something diverges during learning.")
W_nup = get_W(sp_nup)
W_lat = get_W(sp_lat)
self.assertTrue(
np.all(W_nup == W_lat),
"Wrong synaptic weights, something diverges during learning.")
def main():
x = 10
y = 10
steps = 10000
ww = 50
wn = ww ** 2
rr = 3
rw = rr * 2 + 1
nCols = 25
history = []
world = np.array([i for i in xrange(wn)])
world.resize((ww, ww))
binaryWorld = np.zeros((wn,), dtype=np.uint32)
binaryWorld[:wn/2] = 1
np.random.shuffle(binaryWorld)
sp = SpatialPooler(
inputDimensions=(rw ** 2,),
columnDimensions=(nCols,),
potentialRadius=25,
potentialPct=1.0,
numActiveColumnsPerInhArea=1,
synPermActiveInc=0.01,
synPermInactiveDec=0.003,
boostStrength=15.0,
dutyCyclePeriod=5,
)
output = np.zeros((nCols,), dtype=np.uint32)
for _ in xrange(steps):
active = getActive(world, ww, x, y, rr)
assert len(active) == (rw) ** 2, "{}, {}: {}".format(x, y, active)
active = np.array(list(active))
activeInput = binaryWorld[active]
sp.compute(activeInput, True, output)
x, y = getNewLocation(x, y, ww, rr, False)
# Check firing fields
sp.setBoostStrength(0.0)
firingFields = {}
for i in xrange(nCols):
firingFields[i] = np.zeros((ww-rw, ww-rw), dtype=np.uint32)
for i in xrange(0, ww-rw+1):
for j in xrange(0, ww-rw+1):
active = getActive(world, ww, i, j, rr)
active = np.array(list(active))
activeInput = binaryWorld[active]
sp.compute(activeInput, True, output)
for column in list(output.nonzero()):
firingFields[column[0]][i:i+rr, j:j+rr] += 1
for col, ff in firingFields.iteritems():
plt.imshow(ff, cmap="hot", interpolation="nearest")
plt.show()
def testCompatibilityCppPyDirectCall2D(self):
"""Check SP implementations have same behavior with 2D input."""
pySp = PySpatialPooler(
inputDimensions=[121, 1], columnDimensions=[30, 30])
cppSp = CPPSpatialPooler(
inputDimensions=[121, 1], columnDimensions=[30, 30])
data = numpy.zeros([121, 1], dtype=uintType)
for i in xrange(21):
data[i][0] = 1
nCols = 900
d1 = numpy.zeros(nCols, dtype=uintType)
d2 = numpy.zeros(nCols, dtype=uintType)
pySp.compute(data, True, d1) # learn
cppSp.compute(data, True, d2)
d1 = d1.nonzero()[0].tolist()
d2 = d2.nonzero()[0].tolist()
self.assertListEqual(
d1, d2, "SP outputs are not equal: \n%s \n%s" % (str(d1), str(d2)))
def loadNetwork(path):
""" Deserialize the network from the given path
"""
# Create the network structure
(net, layers) = createNetwork()
# Replace algorithm components with loaded (connections etc.)
for key, layer in layers.iteritems():
filename = "%s/%s.tmp" % (path, key)
print 'load', key, filename
with open(filename, "rb") as f:
# AgentStateRegion with ETM instance
if 'L5_TM' in key:
proto = AgentStateRegionProto.read_packed(f)
layers[key] = AgentStateRegion.readFromProto(proto)
# ReinforcementRegion with ETM instance
elif any(name in key for name in ['D1_TM', 'D2_TM']):
proto = ReinforcementRegionProto.read_packed(f)
layers[key] = ReinforcementRegion.readFromProto(proto)
# TemporalPoolerRegion using UnionTemporalPooler (extended Spatial Pooler)
elif 'TP' in key:
proto = MyTemporalPoolerRegionProto.read_packed(f)
layers[key] = MyTemporalPoolerRegion.readFromProto(proto)
elif 'Motor' in key:
proto = MotorRegionProto.read_packed(f)
layers[key] = MRegion.readFromProto(proto)
else:
# MySPRegion - sp cpp instance (L4,L5,D1,D2)
if 'SP' in key:
proto = SpatialPoolerProto.read_packed(f)
instance = SpatialPooler.read(proto)
# MyTMRegion - cpp backtrackingTM instance (L2,L3)
elif any(name in key for name in ['L2_TM', 'L3_TM']):
instance = BacktrackingTMCPP.readFromFile(f)
# ExtendedTemporalMemory - cpp instance (L4,L5,D1,D2)
elif 'TM' in key:
# Read from file - schema.read(file)
proto = ExtendedTemporalMemoryProto.read_packed(f)
# Generate class instance from the schema
instance = ExtendedTemporalMemory.read(proto)
elif 'sensor' in key:
continue
layer.setAlgorithmInstance(instance)
return (net, layers)
def main():
# cluster similar inputs together in SDR space
s = SpatialPooler()
print(type(s))
# powerful sequence memory in SDR space
t = TemporalMemory()
print(type(t))
# computes rolling Gaussian based on raw anomaly scores and then their
# likelihood
a = AnomalyLikelihood()
print(type(a))
# temporally groups active cell sets from TM
u = UnionTemporalPooler()
print(type(u))
# learning pairings of Union representations and labeled classes
c = SDRClassifier()
print(type(c))
def main():
x = 10
y = 10
steps = 10000
world = np.array([i for i in xrange(625)])
world.resize((25, 25))
spInputSize = 21 * 21
sp = SpatialPooler(
inputDimensions=(spInputSize, ),
columnDimensions=(25, ),
potentialRadius=spInputSize,
numActiveColumnsPerInhArea=1,
synPermActiveInc=0.1,
synPermInactiveDec=0.5,
boostStrength=1.0,
)
csFields = generateCenterSurroundFields()
output = np.zeros((25, ), dtype=np.uint32)
for _ in xrange(steps):
active = getActive(world, x, y)
assert len(active) == 25, "{}, {}: {}".format(x, y, active)
activeInput = np.zeros((625, ), dtype=np.uint32)
for v in active:
activeInput[v] = 1
centerSurround = processCenterSurround(csFields, activeInput)
print centerSurround
sp.compute(centerSurround, True, output)
x, y = getNewLocation(x, y, 25, 2, False)
for i in xrange(25):
permanence = np.zeros((spInputSize, ))
sp.getPermanence(i, permanence)
plt.imshow(permanence.reshape((21, 21)),
cmap="hot",
interpolation="nearest")
plt.show()
def HTM_AD(
Data='Test',
vars={'value': ['num']},
prec_param=5,
pooler_out=2024, # Number of columns of the pooler output
cell_col=5, # HTM cells per column
W=72, # Window parameter
W_prim=5, # Local window for anomaly detection likelihood
eps=1e-6, # to Avoid by zero divisions
athreshold=0.95):
"""
This function performs HTM based anomaly detection on a time series provided
:param Data:
:param vars: Possible values: num, tod, weekend
:param prec_param: A parameter that defines how much precision the number encoder has
The encoder precision depends on the variability of the data,
The real precision is computed taking into account both the precision parameter and data std
A high precision might mean a high error at predicting the variable value in noisy variables
:param pooler_out: Number of columns of the pooler output
:param cell_col: HTM cells per column
:param W: Window parameter
:param W_prim: Local window for anomaly detection likelihood
:param eps: to Avoid by zero divisions
:param athreshold: To classify based on anomaly likelihood whether there is an anomaly or not
:return: The Data + 3 columns
Anomaly: indicates the error of within the value predicted by the HTM network
Anomaly_likelihood: indicates the likelihood of the data into being anomalous
Anomaly_flag: classifies the data in anomalous vs non anomalous
"""
if Data == 'Test': # If there is not data available, simply loads the temperature benchmark dataset
# Import data
Data = pd.read_csv('anomaly_API/Data/sample.csv',
parse_dates=True,
index_col='timestamp')
Data = Data.resample('H').bfill().interpolate()
TODE = DateEncoder(timeOfDay=(21, 1))
WENDE = DateEncoder(weekend=21)
var_encoders = set()
# Spatial Pooler Parameters
for x in vars:
for y in vars[x]:
if y == 'num':
exec(
"RDSE_" + x +
" = RandomDistributedScalarEncoder(resolution=Data['" + x +
"'].std()/prec_param)", locals(), globals())
var_encoders.add(Encoder(x, ["RDSE_" + x]))
elif y == 'weekend':
var_encoders.add(Encoder(x, ["WENDE"]))
elif y == 'tod':
var_encoders.add(Encoder(x, ["TODE"]))
else:
return {"error": "Variable encoder type is not recognized "}
encoder_width = 0 # Computes encoder width
for x in var_encoders:
for y in x.encoders:
exec("s = " + y + ".getWidth()", locals(), globals())
encoder_width += s
SP = SpatialPooler(
inputDimensions=encoder_width,
columnDimensions=pooler_out,
potentialPct=0.8,
globalInhibition=True,
numActiveColumnsPerInhArea=pooler_out //
50, # Gets 2% of the total area
boostStrength=1.0,
wrapAround=False)
TM = TemporalMemory(columnDimensions=(pooler_out, ),
cellsPerColumn=cell_col)
Data['Anomaly'] = 0.0
Data['Anomaly_Likelihood'] = 0.0
# Train Spatial Pooler
print("Spatial pooler learning")
start = time.time()
active_columns = np.zeros(pooler_out)
for x in range(len(Data)):
encoder = multiencode(var_encoders, Data, x)
SP.compute(encoder, True, active_columns)
end = time.time()
print(end - start)
# Temporal pooler
print("Temporal pooler learning")
start = time.time()
A_score = np.zeros(len(Data))
for x in range(len(Data)):
encoder = multiencode(var_encoders, Data, x)
SP.compute(encoder, False, active_columns)
col_index = active_columns.nonzero()[0]
TM.compute(col_index, learn=True)
if x > 0:
inter = set(col_index).intersection(Prev_pred_col)
inter_l = len(inter)
active_l = len(col_index)
A_score[x] = 1 - (inter_l / active_l)
Data.iat[x, -2] = A_score[x]
Prev_pred_col = list(
set(x // cell_col for x in TM.getPredictiveCells()))
end = time.time()
print(end - start)
AL_score = np.zeros(len(Data))
# Computes the likelihood of the anomaly
for x in range(len(Data)):
if x > 0:
W_vec = A_score[max(0, x - W):x]
W_prim_vec = A_score[max(0, x - W_prim):x]
AL_score[x] = 1 - 2 * norm.sf(
abs(np.mean(W_vec) - np.mean(W_prim_vec)) /
max(np.std(W_vec), eps))
Data.iat[x, -1] = AL_score[x]
Data['Anomaly_flag'] = athreshold < Data['Anomaly_Likelihood']
return Data
uintType = "uint32" inputDimensions = (1000,1) columnDimensions = (2048,1) inputSize = np.array(inputDimensions).prod() columnNumber = np.array(columnDimensions).prod() inputArray = np.zeros(inputSize, dtype=uintType) for i in range(inputSize): inputArray[i] = random.randrange(2) activeCols = np.zeros(columnNumber, dtype=uintType) sp = SP(inputDimensions, columnDimensions, potentialRadius = int(0.5*inputSize), numActiveColumnsPerInhArea = int(0.02*columnNumber), globalInhibition = True, seed = 1, synPermActiveInc = 0.01, synPermInactiveDec = 0.008 ) # Part 1: # ------- # A column connects to a subset of the input vector (specified # by both the potentialRadius and potentialPct). The overlap score # for a column is the number of connections to the input that become # active when presented with a vector. When learning is 'on' in the SP, # the active connections are reinforced, whereas those inactive are # depressed (according to parameters synPermActiveInc and synPermInactiveDec. # In order for the SP to create a sparse representation of the input, it # will select a small percentage (usually 2%) of its most active columns,
def testInhibition(self):
"""
Test if the firing number of coincidences after inhibition
equals spatial pooler numActiveColumnsPerInhArea.
"""
# Miscellaneous variables:
# n, w: n, w of encoders
# inputLen: Length of binary input
# synPermConnected: Spatial pooler synPermConnected
# synPermActiveInc: Spatial pooler synPermActiveInc
# connectPct: Initial connect percentage of permanences
# columnDimensions: Number of spatial pooler coincidences
# numActiveColumnsPerInhArea: Spatial pooler numActiveColumnsPerInhArea
# stimulusThreshold: Spatial pooler stimulusThreshold
# spSeed: Spatial pooler for initial permanences
# stimulusThresholdInh: Parameter for inhibition, default value 0.00001
# kDutyCycleFactor: kDutyCycleFactor for dutyCycleTieBreaker in
# Inhibition
# spVerbosity: Verbosity to print other sp initial parameters
# testIter: Testing iterations
n = 100
w = 15
inputLen = 300
columnDimensions = 2048
numActiveColumnsPerInhArea = 40
stimulusThreshold = 0
spSeed = 1956
stimulusThresholdInh = 0.00001
kDutyCycleFactor = 0.01
spVerbosity = 0
testIter = 100
spTest = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, inputLen),
potentialRadius=inputLen / 2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
stimulusThreshold=stimulusThreshold,
seed=spSeed
)
initialPermanence = spTest._initialPermanence()
spTest._masterPotentialM, spTest._masterPermanenceM = (
spTest._makeMasterCoincidences(spTest.numCloneMasters,
spTest._coincRFShape,
spTest.potentialPct,
initialPermanence,
spTest.random))
spTest._updateInhibitionObj()
boostFactors = numpy.ones(columnDimensions)
for i in range(testIter):
spTest._iterNum = i
# random binary input
input_ = numpy.zeros((1, inputLen))
nonzero = numpy.random.random(inputLen)
input_[0][numpy.where (nonzero < float(w)/float(n))] = 1
# overlap step
spTest._computeOverlapsFP(input_,
stimulusThreshold=spTest.stimulusThreshold)
spTest._overlaps *= boostFactors
onCellIndices = numpy.where(spTest._overlaps > 0)
spTest._onCells.fill(0)
spTest._onCells[onCellIndices] = 1
denseOn = spTest._onCells
# update _dutyCycleBeforeInh
spTest.dutyCyclePeriod = min(i + 1, 1000)
spTest._dutyCycleBeforeInh = (
(spTest.dutyCyclePeriod - 1) *
spTest._dutyCycleBeforeInh +denseOn) / spTest.dutyCyclePeriod
dutyCycleTieBreaker = spTest._dutyCycleAfterInh.copy()
dutyCycleTieBreaker *= kDutyCycleFactor
# inhibition step
numOn = spTest._inhibitionObj.compute(
spTest._overlaps + dutyCycleTieBreaker, spTest._onCellIndices,
stimulusThresholdInh, # stimulusThresholdInh
max(spTest._overlaps)/1000, # addToWinners
)
# update _dutyCycleAfterInh
spTest._onCells.fill(0)
onCellIndices = spTest._onCellIndices[0:numOn]
spTest._onCells[onCellIndices] = 1
denseOn = spTest._onCells
spTest._dutyCycleAfterInh = (((spTest.dutyCyclePeriod-1) *
spTest._dutyCycleAfterInh + denseOn) /
spTest.dutyCyclePeriod)
# learning step
spTest._adaptSynapses(onCellIndices, [], input_)
# update boostFactor
spTest._updateBoostFactors()
boostFactors = spTest._firingBoostFactors
# update dutyCycle and boost
if ((spTest._iterNum+1) % 50) == 0:
spTest._updateInhibitionObj()
spTest._updateMinDutyCycles(
spTest._dutyCycleBeforeInh,
spTest.minPctDutyCycleBeforeInh,
spTest._minDutyCycleBeforeInh)
spTest._updateMinDutyCycles(
spTest._dutyCycleAfterInh,
spTest.minPctDutyCycleAfterInh,
spTest._minDutyCycleAfterInh)
# test numOn and spTest.numActiveColumnsPerInhArea
self.assertEqual(numOn, spTest.numActiveColumnsPerInhArea,
"Error at input %s, actual numOn are: %i, "
"numActivePerInhAre is: %s" % (
i, numOn, numActiveColumnsPerInhArea))
def _runLearnInference(self,
n=30,
w=15,
columnDimensions=2048,
numActiveColumnsPerInhArea=40,
spSeed=1951,
spVerbosity=0,
numTrainingRecords=100,
seed=42):
# Instantiate two identical spatial pooler. One will be used only for
# learning. The other will be trained with identical records, but with
# random inference calls thrown in
spLearnOnly = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, n),
potentialRadius=n / 2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
seed=spSeed,
synPermInactiveDec=0.01,
synPermActiveInc=0.2,
synPermConnected=0.11,
)
spLearnInfer = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, n),
potentialRadius=n / 2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
seed=spSeed,
synPermInactiveDec=0.01,
synPermActiveInc=0.2,
synPermConnected=0.11,
)
random.seed(seed)
np.random.seed(seed)
# Build up training set with numTrainingRecords patterns
inputs = [] # holds post-encoded input patterns
for i in xrange(numTrainingRecords):
inputVector = np.zeros(n, dtype=realDType)
inputVector[random.sample(xrange(n), w)] = 1
inputs.append(inputVector)
# Train each SP with identical inputs
startTime = time.time()
random.seed(seed)
np.random.seed(seed)
for i in xrange(numTrainingRecords):
if spVerbosity > 0:
print "Input #%d" % i
# TODO: See https://github.com/numenta/nupic/issues/2072
encodedInput = inputs[i]
decodedOutput = np.zeros(columnDimensions)
spLearnOnly.compute(encodedInput,
learn=True,
activeArray=decodedOutput)
random.seed(seed)
np.random.seed(seed)
for i in xrange(numTrainingRecords):
if spVerbosity > 0:
print "Input #%d" % i
# TODO: See https://github.com/numenta/nupic/issues/2072
encodedInput = inputs[i]
decodedOutput = np.zeros(columnDimensions)
spLearnInfer.compute(encodedInput,
learn=True,
activeArray=decodedOutput)
print "\nElapsed time: %.2f seconds\n" % (time.time() - startTime)
# Test that both SP"s are identical by checking learning stats
# A more in depth test would check all the coincidences, duty cycles, etc.
# ala tpDiff
# Edit: spDiff has been written as an in depth tester of the spatial pooler
learnOnlyStats = spLearnOnly.getLearningStats()
learnInferStats = spLearnInfer.getLearningStats()
success = True
# Check that the two spatial poolers are equivalent after the same training.
success = success and spDiff(spLearnInfer, spLearnOnly)
self.assertTrue(success)
# Make sure that the pickled and loaded SPs are equivalent.
spPickle = pickle.dumps(spLearnOnly, protocol=0)
spLearnOnlyLoaded = pickle.loads(spPickle)
success = success and spDiff(spLearnOnly, spLearnOnlyLoaded)
self.assertTrue(success)
for k in learnOnlyStats.keys():
if learnOnlyStats[k] != learnInferStats[k]:
success = False
print "Stat", k, "is different:", learnOnlyStats[
k], learnInferStats[k]
self.assertTrue(success)
if success:
print "Test succeeded"
def runHotgym(numRecords):
with open(_PARAMS_PATH, "r") as f:
modelParams = yaml.safe_load(f)["modelParams"]
enParams = modelParams["sensorParams"]["encoders"]
spParams = modelParams["spParams"]
tmParams = modelParams["tmParams"]
timeOfDayEncoder = DateEncoder(
timeOfDay=enParams["timestamp_timeOfDay"]["timeOfDay"])
weekendEncoder = DateEncoder(
weekend=enParams["timestamp_weekend"]["weekend"])
scalarEncoder = RandomDistributedScalarEncoder(
enParams["consumption"]["resolution"])
encodingWidth = (timeOfDayEncoder.getWidth() + weekendEncoder.getWidth() +
scalarEncoder.getWidth())
sp = SpatialPooler(
# How large the input encoding will be.
inputDimensions=(encodingWidth, ),
# How many mini-columns will be in the Spatial Pooler.
columnDimensions=(spParams["columnCount"], ),
# What percent of the columns"s receptive field is available for potential
# synapses?
potentialPct=spParams["potentialPct"],
# Potential radius should be set to the input size if there is global
# inhibition.
potentialRadius=encodingWidth,
# This means that the input space has no topology.
globalInhibition=spParams["globalInhibition"],
localAreaDensity=spParams["localAreaDensity"],
# Roughly 2%, giving that there is only one inhibition area because we have
# turned on globalInhibition (40 / 2048 = 0.0195)
numActiveColumnsPerInhArea=spParams["numActiveColumnsPerInhArea"],
# How quickly synapses grow and degrade.
synPermInactiveDec=spParams["synPermInactiveDec"],
synPermActiveInc=spParams["synPermActiveInc"],
synPermConnected=spParams["synPermConnected"],
# boostStrength controls the strength of boosting. Boosting encourages
# efficient usage of SP columns.
boostStrength=spParams["boostStrength"],
# Random number generator seed.
seed=spParams["seed"],
# TODO: is this useful?
# Determines if inputs at the beginning and end of an input dimension should
# be considered neighbors when mapping columns to inputs.
wrapAround=True)
tm = TemporalMemory(
# Must be the same dimensions as the SP
columnDimensions=(tmParams["columnCount"], ),
# How many cells in each mini-column.
cellsPerColumn=tmParams["cellsPerColumn"],
# A segment is active if it has >= activationThreshold connected synapses
# that are active due to infActiveState
activationThreshold=tmParams["activationThreshold"],
initialPermanence=tmParams["initialPerm"],
# TODO: This comes from the SP params, is this normal
connectedPermanence=spParams["synPermConnected"],
# Minimum number of active synapses for a segment to be considered during
# search for the best-matching segments.
minThreshold=tmParams["minThreshold"],
# The max number of synapses added to a segment during learning
maxNewSynapseCount=tmParams["newSynapseCount"],
permanenceIncrement=tmParams["permanenceInc"],
permanenceDecrement=tmParams["permanenceDec"],
predictedSegmentDecrement=0.0,
maxSegmentsPerCell=tmParams["maxSegmentsPerCell"],
maxSynapsesPerSegment=tmParams["maxSynapsesPerSegment"],
seed=tmParams["seed"])
classifier = SDRClassifierFactory.create()
results = []
with open(_INPUT_FILE_PATH, "r") as fin:
reader = csv.reader(fin)
headers = reader.next()
reader.next()
reader.next()
for count, record in enumerate(reader):
if count >= numRecords: break
# Convert data string into Python date object.
dateString = datetime.datetime.strptime(record[0],
"%m/%d/%y %H:%M")
# Convert data value string into float.
consumption = float(record[1])
# To encode, we need to provide zero-filled numpy arrays for the encoders
# to populate.
timeOfDayBits = numpy.zeros(timeOfDayEncoder.getWidth())
weekendBits = numpy.zeros(weekendEncoder.getWidth())
consumptionBits = numpy.zeros(scalarEncoder.getWidth())
# Now we call the encoders create bit representations for each value.
timeOfDayEncoder.encodeIntoArray(dateString, timeOfDayBits)
weekendEncoder.encodeIntoArray(dateString, weekendBits)
scalarEncoder.encodeIntoArray(consumption, consumptionBits)
# Concatenate all these encodings into one large encoding for Spatial
# Pooling.
encoding = numpy.concatenate(
[timeOfDayBits, weekendBits, consumptionBits])
# Create an array to represent active columns, all initially zero. This
# will be populated by the compute method below. It must have the same
# dimensions as the Spatial Pooler.
activeColumns = numpy.zeros(spParams["columnCount"])
# Execute Spatial Pooling algorithm over input space.
sp.compute(encoding, True, activeColumns)
activeColumnIndices = numpy.nonzero(activeColumns)[0]
# Execute Temporal Memory algorithm over active mini-columns.
tm.compute(activeColumnIndices, learn=True)
activeCells = tm.getActiveCells()
# Get the bucket info for this input value for classification.
bucketIdx = scalarEncoder.getBucketIndices(consumption)[0]
# Run classifier to translate active cells back to scalar value.
classifierResult = classifier.compute(recordNum=count,
patternNZ=activeCells,
classification={
"bucketIdx": bucketIdx,
"actValue": consumption
},
learn=True,
infer=True)
# Print the best prediction for 1 step out.
oneStepConfidence, oneStep = sorted(zip(
classifierResult[1], classifierResult["actualValues"]),
reverse=True)[0]
print("1-step: {:16} ({:4.4}%)".format(oneStep,
oneStepConfidence * 100))
results.append([oneStep, oneStepConfidence * 100, None, None])
return results
def initialize(self):
# Keep track of value range for spatial anomaly detection.
self.minVal = None
self.maxVal = None
# Time of day encoder
self.timeOfDayEncoder = DateEncoder(timeOfDay=(21, 9.49),
name='time_enc')
# RDSE encoder for the time series value.
minResolution = 0.001
rangePadding = abs(self.inputMax - self.inputMin) * 0.2
minVal = self.inputMin - rangePadding
maxVal = self.inputMax + rangePadding
numBuckets = 130
resolution = max(minResolution, (maxVal - minVal) / numBuckets)
self.value_enc = RandomDistributedScalarEncoder(resolution=resolution,
name='value_rdse')
# Spatial Pooler.
encodingWidth = self.timeOfDayEncoder.getWidth(
) + self.value_enc.getWidth()
self.sp = SpatialPooler(
inputDimensions=(encodingWidth, ),
columnDimensions=(2048, ),
potentialPct=0.8,
potentialRadius=encodingWidth,
globalInhibition=1,
numActiveColumnsPerInhArea=40,
synPermInactiveDec=0.0005,
synPermActiveInc=0.003,
synPermConnected=0.2,
boostStrength=0.0,
seed=1956,
wrapAround=True,
)
self.tm = TemporalMemory(
columnDimensions=(2048, ),
cellsPerColumn=32,
activationThreshold=20,
initialPermanence=.5, # Increased to connectedPermanence.
connectedPermanence=.5,
minThreshold=13,
maxNewSynapseCount=31,
permanenceIncrement=0.04,
permanenceDecrement=0.008,
predictedSegmentDecrement=0.001,
maxSegmentsPerCell=128,
maxSynapsesPerSegment=
128, # Changed meaning. Also see connections.topology[2]
seed=1993,
)
# Initialize the anomaly likelihood object
numentaLearningPeriod = int(math.floor(self.probationaryPeriod / 2.0))
self.anomalyLikelihood = anomaly_likelihood.AnomalyLikelihood(
learningPeriod=numentaLearningPeriod,
estimationSamples=self.probationaryPeriod - numentaLearningPeriod,
reestimationPeriod=100,
)
self.age = 0
def main(argv):
args, _ = parse_argv()
data_set = args.data_set
num_data_points = args.num_data_points
sp_type = args.pooler_type
num_epochs = args.num_epochs
batch_size = args.batch_size
experiment_id = args.experiment_id
seed = args.seed
the_scripts_path = os.path.dirname(os.path.realpath(__file__)) # script directory
sp_params_dict = json.load(open(the_scripts_path + "/params.json"))
if args.sp_params is not None:
sp_params = sp_params_dict[sp_type][args.sp_params]
else:
sp_params = sp_params_dict[sp_type][data_set]
sp_params["seed"] = seed
if experiment_id is None:
experiment_id = random_id(5)
path = the_scripts_path + "/../results/{}_pooler_{}_{}/".format(sp_type, data_set,experiment_id)
os.makedirs(os.path.dirname(path))
print(
"Experiment directory:\n\"{}\"\n"
.format(path))
X, _, X_test, _ = load_data(data_set, num_inputs = num_data_points)
n, m = get_shape(sp_params)
X = X[:,:num_data_points]
d = X.shape[1]
results = {
"inputs" : [],
"outputs" : [],
"feedforward" : []}
####################################################
#
# Old Spatial Pooler
#
####################################################
if sp_type == "ordinary":
pooler = OldSpatialPooler(**sp_params)
print(
"Training ordinary pooler:\n")
# "Fit" the model to the training data
for epoch in range(num_epochs):
Y = np.zeros((n,d))
perm = np.random.permutation(d)
for t in range(d):
sys.stdout.flush()
sys.stdout.write(
"\r{}/{} {}/{}"
.format(num_epochs, epoch + 1, d, t + 1))
x = X[:,perm[t]]
y = Y[:, t]
pooler.compute(x, True, y)
results["inputs"].append(X)
results["outputs"].append(Y)
results["feedforward"].append(get_permanence_vals(pooler))
####################################################
#
# New Spatial Pooler with
# learned lateral inhibitory connections
#
#####################################################
elif sp_type == "lateral":
pooler = LateralPooler(**sp_params)
sys.stdout.write(
"Training dynamic lateral pooler:\n")
collect_feedforward = ModelInspector(lambda pooler: pooler.feedforward.copy(), on_batch = False )
# collect_lateral = ModelInspector(lambda pooler: pooler.inhibitory.copy(), on_batch = False )
training_log = OutputCollector()
print_training_status = Logger()
# "Fit" the model to the training datasets
pooler.fit(X, batch_size=batch_size, num_epochs=num_epochs, initial_epoch=0, callbacks=[collect_feedforward, training_log, print_training_status])
results["inputs"] = training_log.get_inputs()
results["outputs"] = training_log.get_outputs()
results["feedforward"] = collect_feedforward.get_results()
# results["lateral"] = collect_lateral.get_results()
dump_dict(path, sp_params)
dump_results(path, results)
print(
"\nDone.\n")
class Entity():
def __init__(self, columnCount, InputEncoderParams, toL4ConnectorParamsI,
toL4ConnectorParamsII, toL5ConnectorParams,
toD1ConnectorParams, toD2ConnectorParams, L4Params, L5Params,
k, D1Params, D2Params):
self.columnCount = columnCount
self.toL4ConnectorParamsI = toL4ConnectorParamsI
self.toL4ConnectorParamsII = toL4ConnectorParamsII
self.toL5ConnectorParams = toL5ConnectorParams
self.toD1ConnectorParams = toD1ConnectorParams
self.toD2ConnectorParams = toD2ConnectorParams
self.L4Params = L4Params
self.L5Params = L5Params
self.k = k
self.D1Params = D1Params
self.D2Params = D2Params
self.learning = False
#encoder
from nupic.encoders import MultiEncoder
self.InputEncoder = MultiEncoder()
self.InputEncoder.addMultipleEncoders(InputEncoderParams)
print "Encoder Online"
#spatialPoolers
from nupic.algorithms.spatial_pooler import SpatialPooler
self.toL4ConnectorI = SpatialPooler(
inputDimensions=(toL4ConnectorParamsI["inputDimensions"], ),
columnDimensions=(columnCount, ),
potentialPct=toL4ConnectorParamsI["potentialPct"],
globalInhibition=toL4ConnectorParamsI["globalInhibition"],
localAreaDensity=toL4ConnectorParamsI["localAreaDensity"],
numActiveColumnsPerInhArea=toL4ConnectorParamsI[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL4ConnectorParamsI["synPermInactiveDec"],
synPermActiveInc=toL4ConnectorParamsI["synPermActiveInc"],
synPermConnected=toL4ConnectorParamsI["synPermConnected"],
boostStrength=toL4ConnectorParamsI["boostStrength"],
seed=toL4ConnectorParamsI["seed"],
wrapAround=toL4ConnectorParamsI["wrapAround"]) #this part sucks
self.toL4ConnectorII = SpatialPooler(
inputDimensions=(columnCount * 3, ),
columnDimensions=(columnCount, ),
potentialPct=toL4ConnectorParamsII["potentialPct"],
globalInhibition=toL4ConnectorParamsII["globalInhibition"],
localAreaDensity=toL4ConnectorParamsII["localAreaDensity"],
numActiveColumnsPerInhArea=toL4ConnectorParamsII[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL4ConnectorParamsII["synPermInactiveDec"],
synPermActiveInc=toL4ConnectorParamsII["synPermActiveInc"],
synPermConnected=toL4ConnectorParamsII["synPermConnected"],
boostStrength=toL4ConnectorParamsII["boostStrength"],
seed=toL4ConnectorParamsII["seed"],
wrapAround=toL4ConnectorParamsII["wrapAround"])
print "toL4Connector Online"
self.toL5Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toL5ConnectorParams["potentialPct"],
globalInhibition=toL5ConnectorParams["globalInhibition"],
localAreaDensity=toL5ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toL5ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL5ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toL5ConnectorParams["synPermActiveInc"],
synPermConnected=toL5ConnectorParams["synPermConnected"],
boostStrength=toL5ConnectorParams["boostStrength"],
seed=toL5ConnectorParams["seed"],
wrapAround=toL5ConnectorParams["wrapAround"])
print "toL5Connector Online"
self.toD1Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toD1ConnectorParams["potentialPct"],
globalInhibition=toD1ConnectorParams["globalInhibition"],
localAreaDensity=toD1ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toD1ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toD1ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toD1ConnectorParams["synPermActiveInc"],
synPermConnected=toD1ConnectorParams["synPermConnected"],
boostStrength=toD1ConnectorParams["boostStrength"],
seed=toD1ConnectorParams["seed"],
wrapAround=toD1ConnectorParams["wrapAround"])
print "toD1Connector Online"
self.toD2Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toD2ConnectorParams["potentialPct"],
globalInhibition=toD2ConnectorParams["globalInhibition"],
localAreaDensity=toD2ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toD2ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toD2ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toD2ConnectorParams["synPermActiveInc"],
synPermConnected=toD2ConnectorParams["synPermConnected"],
boostStrength=toD2ConnectorParams["boostStrength"],
seed=toD2ConnectorParams["seed"],
wrapAround=toD2ConnectorParams["wrapAround"])
print "toD2Connector Online"
#HTM Layers
from nupic.algorithms.temporal_memory import TemporalMemory
self.L4ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.L4 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
)
print "L4 Online"
self.L5ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.L5 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
)
print "L5 Online"
self.D1ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.D1 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
initialPermanence=0.21,
connectedPermanence=0.5,
)
print "D1 Online"
self.D2ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.D2 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
initialPermanence=0.21,
connectedPermanence=0.5,
)
print "D2 Online"
def encode_input(sine1, sine2, angularSpeed1, angularSpeed2,
efferenceCopy):
return self.InputEncoder.encode({
"sine1": sine1,
"sine2": sine2,
"angularSpeed1": angularSpeed1,
"angularSpeed2": angularSpeed2,
"efferenceCopy": efferenceCopy
})
def reset(self):
self.action = 0
self.L4.reset()
self.L5.reset()
self.D1.reset()
self.D2.reset()
def mimic(self, observation, action):
#mimicking only requires remembering the given obs-act pattern,thus the striatum is neglected in this func
self.learning = True
self.action = action
encodedInput = self.encode_input(observation[0], observation[2],
observation[4], observation[5],
str(action))
self.toL4ConnectorI.compute(encodedInput, self.learning,
self.L4ActiveColumns)
self.L4.compute(self.L4ActiveColumns, learn=self.learning)
L4activeColumnIndices = numpy.nonzero(self.L4ActiveColumns)[0]
L5Temp = numpy.zeros(self.columnCount, dtype=int)
for column in L4activeColumnIndices:
L5Temp[column] = 1
self.toL5Connector.compute(L5Temp, self.learning, self.L5ActiveColumns)
self.L5.compute(self.L5ActiveColumns, learn=self.learning)
L5activeColumnIndices = numpy.nonzero(self.L5ActiveColumns)[0]
#no action generation is needed in this func
def learn(self, env, observation, expectedReaction):
#We humans learn by trial and error,so does an AI agent.For neural networks,they have BP,but HTM does not
#have a clear way to reinforcement learn(Where to feed in rewards?).Here I try to do something new.
self.learning = False #...trial
encodedInput = self.encode_input(observation[0], observation[2],
observation[4], observation[5],
str(self.action))
self.toL4ConnectorI.compute(encodedInput, self.learning,
self.L4ActiveColumns)
L4Temp = numpy.zeros(
self.columnCount * 3, dtype=int
) #ready to receive D1's disinhibition and D2's inhibition
L4activeColumnIndices = numpy.nonzero(self.L4ActiveColumns)[0]
for column in L4activeColumnIndices:
L4Temp[int(column) * 3] = 1
D1ActiveColumnsIndices = numpy.nonzero(self.D1ActiveColumns)[0]
for column in D1ActiveColumnsIndices:
L4Temp[int(column) * 3 + 1] = 1
D2ActiveColumnsIndices = numpy.nonzero(self.D2ActiveColumns)[0]
for i in range(self.columnCount - 1):
L4Temp[i * 3 + 2] = 1
for column in D2ActiveColumnsIndices: #achieve inhibition in this way
L4Temp[i * 3 + 2] = 0
self.toL4ConnectorII.compute(L4Temp, self.learning,
self.L4ActiveColumns)
self.L4.compute(self.L4ActiveColumns, learn=self.learning)
L4activeColumnIndices = numpy.nonzero(self.L4ActiveColumns)[0]
L5Temp = numpy.zeros(self.columnCount, dtype=int)
for column in L4activeColumnIndices:
L5Temp[column] = 1
self.toL5Connector.compute(L5Temp, self.learning, self.L5ActiveColumns)
self.L5.compute(self.L5ActiveColumns, learn=self.learning)
L5activeColumnIndices = numpy.nonzero(self.L5ActiveColumns)[0]
#Action Generation
p = 84 #there are 84 bits in the SDR representing the action fed in the agent,this is the "Efference Copy"
count0 = 0
count1 = 0
count2 = 0
for activeIndice in L5activeColumnIndices:
convertedIndice = (activeIndice + 1) * 1126 / columnCount
if convertedIndice <= 1126 - p / 4 and convertedIndice > 1126 - p / 2:
count2 = count2 + 1
if convertedIndice <= 1126 - p / 2 and convertedIndice > 1126 - 3 * p / 4:
count1 = count1 + 1
if convertedIndice <= 1126 - 3 * p / 4 and convertedIndice > 1126 - p:
count0 = count0 + 1
if count2 == max(count0, count1, count2):
self.action = 2
if count1 == max(count0, count1, count2):
self.action = 1
if count0 == max(count0, count1, count2):
self.action = 0
#...and error
if self.action == expectedReaction:
reward = 0.1
else:
reward = -0.1
self.D1.setConnectedPermanence(self.D1.getConnectedPermanence() *
(self.k**(-reward))) #reward
self.D2.setConnectedPermanence(self.D2.getConnectedPermanence() *
(self.k**reward)) #punishment
#Learn to correct mistakes(remember what's right and whats' wrong)
self.learning = True
DTemp = numpy.zeros(self.columnCount, dtype=int)
for column in L5activeColumnIndices:
DTemp[column] = 1
self.toD1Connector.compute(DTemp, self.learning, self.D1ActiveColumns)
self.toD2Connector.compute(DTemp, self.learning, self.D2ActiveColumns)
self.D1.compute(self.D1ActiveColumns, learn=self.learning)
self.D2.compute(self.D2ActiveColumns, learn=self.learning)
return reward
def react(self, observation):
self.learning = False
encodedInput = self.encode_input(observation[0], observation[2],
observation[4], observation[5],
str(self.action))
self.toL4ConnectorI.compute(encodedInput, self.learning,
self.L4ActiveColumns)
L4Temp = numpy.zeros(self.columnCount * 3, dtype=int)
L4activeColumnIndices = numpy.nonzero(self.L4ActiveColumns)[0]
for column in L4activeColumnIndices:
L4Temp[int(column) * 3] = 1
D1ActiveColumnsIndices = numpy.nonzero(self.D1ActiveColumns)[0]
for column in D1ActiveColumnsIndices:
L4Temp[int(column) * 3 + 1] = 1
D2ActiveColumnsIndices = numpy.nonzero(self.D2ActiveColumns)[0]
for i in range(self.columnCount - 1):
L4Temp[i * 3 + 2] = 1
for column in D2ActiveColumnsIndices:
L4Temp[i * 3 + 2] = 0
self.toL4ConnectorII.compute(L4Temp, self.learning,
self.L4ActiveColumns)
self.L4.compute(self.L4ActiveColumns, learn=self.learning)
L4activeColumnIndices = numpy.nonzero(self.L4ActiveColumns)[0]
L5Temp = numpy.zeros(self.columnCount, dtype=int)
for column in L4activeColumnIndices:
L5Temp[column] = 1
self.toL5Connector.compute(L5Temp, self.learning, self.L5ActiveColumns)
self.L5.compute(self.L5ActiveColumns, learn=self.learning)
L5activeColumnIndices = numpy.nonzero(self.L5ActiveColumns)[0]
p = 84
count0 = 0
count1 = 0
count2 = 0
for activeIndice in L5activeColumnIndices:
convertedIndice = (activeIndice + 1) * 1126 / columnCount
if convertedIndice <= 1126 - p / 4 and convertedIndice > 1126 - p / 2:
count2 = count2 + 1
if convertedIndice <= 1126 - p / 2 and convertedIndice > 1126 - 3 * p / 4:
count1 = count1 + 1
if convertedIndice <= 1126 - 3 * p / 4 and convertedIndice > 1126 - p:
count0 = count0 + 1
if count2 == max(count0, count1, count2):
self.action = 2
if count1 == max(count0, count1, count2):
self.action = 1
if count0 == max(count0, count1, count2):
self.action = 0
return self.action
def runHotgym(numRecords):
with open(_PARAMS_PATH, "r") as f:
modelParams = yaml.safe_load(f)["modelParams"]
enParams = modelParams["sensorParams"]["encoders"]
spParams = modelParams["spParams"]
tmParams = modelParams["tmParams"]
timeOfDayEncoder = DateEncoder(
timeOfDay=enParams["timestamp_timeOfDay"]["timeOfDay"])
weekendEncoder = DateEncoder(
weekend=enParams["timestamp_weekend"]["weekend"])
scalarEncoder = RandomDistributedScalarEncoder(
enParams["consumption"]["resolution"])
encodingWidth = (timeOfDayEncoder.getWidth()
+ weekendEncoder.getWidth()
+ scalarEncoder.getWidth())
sp = SpatialPooler(
inputDimensions=(encodingWidth,),
columnDimensions=(spParams["columnCount"],),
potentialPct=spParams["potentialPct"],
potentialRadius=encodingWidth,
globalInhibition=spParams["globalInhibition"],
localAreaDensity=spParams["localAreaDensity"],
numActiveColumnsPerInhArea=spParams["numActiveColumnsPerInhArea"],
synPermInactiveDec=spParams["synPermInactiveDec"],
synPermActiveInc=spParams["synPermActiveInc"],
synPermConnected=spParams["synPermConnected"],
boostStrength=spParams["boostStrength"],
seed=spParams["seed"],
wrapAround=True
)
tm = TemporalMemory(
columnDimensions=(tmParams["columnCount"],),
cellsPerColumn=tmParams["cellsPerColumn"],
activationThreshold=tmParams["activationThreshold"],
initialPermanence=tmParams["initialPerm"],
connectedPermanence=spParams["synPermConnected"],
minThreshold=tmParams["minThreshold"],
maxNewSynapseCount=tmParams["newSynapseCount"],
permanenceIncrement=tmParams["permanenceInc"],
permanenceDecrement=tmParams["permanenceDec"],
predictedSegmentDecrement=0.0,
maxSegmentsPerCell=tmParams["maxSegmentsPerCell"],
maxSynapsesPerSegment=tmParams["maxSynapsesPerSegment"],
seed=tmParams["seed"]
)
classifier = SDRClassifierFactory.create()
results = []
with open(_INPUT_FILE_PATH, "r") as fin:
reader = csv.reader(fin)
headers = reader.next()
reader.next()
reader.next()
for count, record in enumerate(reader):
if count >= numRecords: break
# Convert data string into Python date object.
dateString = datetime.datetime.strptime(record[0], "%m/%d/%y %H:%M")
# Convert data value string into float.
consumption = float(record[1])
# To encode, we need to provide zero-filled numpy arrays for the encoders
# to populate.
timeOfDayBits = numpy.zeros(timeOfDayEncoder.getWidth())
weekendBits = numpy.zeros(weekendEncoder.getWidth())
consumptionBits = numpy.zeros(scalarEncoder.getWidth())
# Now we call the encoders to create bit representations for each value.
timeOfDayEncoder.encodeIntoArray(dateString, timeOfDayBits)
weekendEncoder.encodeIntoArray(dateString, weekendBits)
scalarEncoder.encodeIntoArray(consumption, consumptionBits)
# Concatenate all these encodings into one large encoding for Spatial
# Pooling.
encoding = numpy.concatenate(
[timeOfDayBits, weekendBits, consumptionBits]
)
# Create an array to represent active columns, all initially zero. This
# will be populated by the compute method below. It must have the same
# dimensions as the Spatial Pooler.
activeColumns = numpy.zeros(spParams["columnCount"])
# Execute Spatial Pooling algorithm over input space.
sp.compute(encoding, True, activeColumns)
activeColumnIndices = numpy.nonzero(activeColumns)[0]
# Execute Temporal Memory algorithm over active mini-columns.
tm.compute(activeColumnIndices, learn=True)
activeCells = tm.getActiveCells()
# Get the bucket info for this input value for classification.
bucketIdx = scalarEncoder.getBucketIndices(consumption)[0]
# Run classifier to translate active cells back to scalar value.
classifierResult = classifier.compute(
recordNum=count,
patternNZ=activeCells,
classification={
"bucketIdx": bucketIdx,
"actValue": consumption
},
learn=True,
infer=True
)
# Print the best prediction for 1 step out.
oneStepConfidence, oneStep = sorted(
zip(classifierResult[1], classifierResult["actualValues"]),
reverse=True
)[0]
print("1-step: {:16} ({:4.4}%)".format(oneStep, oneStepConfidence * 100))
results.append([oneStep, oneStepConfidence * 100, None, None])
return results
now = datetime.datetime.strptime(lines, "%Y-%m-%d %H:%M:%S")
print "now = ", de.encode(now)
cpt += 1
categories = ('info', 'error', 'warning')
encoder = CategoryEncoder(w=3, categoryList=categories, forced=True)
info = encoder.encode("info")
error = encoder.encode("error")
warning = encoder.encode("warning")
#print "info = ", info
#print "error = ", error
#print "warning = ", warning
sp = SpatialPooler(inputDimensions=(len(info), ),
columnDimensions=(3, ),
potentialRadius=15,
numActiveColumnsPerInhArea=1,
globalInhibition=True,
synPermActiveInc=0.03,
potentialPct=1.0)
import numpy
for column in xrange(3):
connected = numpy.zeros((len(info), ), dtype="int")
sp.getConnectedSynapses(column, connected)
print connected
output = numpy.zeros((3, ), dtype="int")
sp.compute(info, learn=True, activeArray=output)
print output
output = numpy.zeros((3, ), dtype="int")
sp.compute(error, learn=True, activeArray=output)
def testInhibition(self):
"""
Test if the firing number of coincidences after inhibition
equals spatial pooler numActiveColumnsPerInhArea.
"""
# Miscellaneous variables:
# n, w: n, w of encoders
# inputLen: Length of binary input
# synPermConnected: Spatial pooler synPermConnected
# synPermActiveInc: Spatial pooler synPermActiveInc
# connectPct: Initial connect percentage of permanences
# columnDimensions: Number of spatial pooler coincidences
# numActiveColumnsPerInhArea: Spatial pooler numActiveColumnsPerInhArea
# stimulusThreshold: Spatial pooler stimulusThreshold
# spSeed: Spatial pooler for initial permanences
# stimulusThresholdInh: Parameter for inhibition, default value 0.00001
# kDutyCycleFactor: kDutyCycleFactor for dutyCycleTieBreaker in
# Inhibition
# spVerbosity: Verbosity to print other sp initial parameters
# testIter: Testing iterations
n = 100
w = 15
inputLen = 300
columnDimensions = 2048
numActiveColumnsPerInhArea = 40
stimulusThreshold = 0
spSeed = 1956
stimulusThresholdInh = 0.00001
kDutyCycleFactor = 0.01
spVerbosity = 0
testIter = 100
spTest = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, inputLen),
potentialRadius=inputLen / 2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
stimulusThreshold=stimulusThreshold,
seed=spSeed)
initialPermanence = spTest._initialPermanence()
spTest._masterPotentialM, spTest._masterPermanenceM = (
spTest._makeMasterCoincidences(spTest.numCloneMasters,
spTest._coincRFShape,
spTest.potentialPct,
initialPermanence, spTest.random))
spTest._updateInhibitionObj()
boostFactors = numpy.ones(columnDimensions)
for i in range(testIter):
spTest._iterNum = i
# random binary input
input_ = numpy.zeros((1, inputLen))
nonzero = numpy.random.random(inputLen)
input_[0][numpy.where(nonzero < float(w) / float(n))] = 1
# overlap step
spTest._computeOverlapsFP(
input_, stimulusThreshold=spTest.stimulusThreshold)
spTest._overlaps *= boostFactors
onCellIndices = numpy.where(spTest._overlaps > 0)
spTest._onCells.fill(0)
spTest._onCells[onCellIndices] = 1
denseOn = spTest._onCells
# update _dutyCycleBeforeInh
spTest.dutyCyclePeriod = min(i + 1, 1000)
spTest._dutyCycleBeforeInh = (
(spTest.dutyCyclePeriod - 1) * spTest._dutyCycleBeforeInh +
denseOn) / spTest.dutyCyclePeriod
dutyCycleTieBreaker = spTest._dutyCycleAfterInh.copy()
dutyCycleTieBreaker *= kDutyCycleFactor
# inhibition step
numOn = spTest._inhibitionObj.compute(
spTest._overlaps + dutyCycleTieBreaker,
spTest._onCellIndices,
stimulusThresholdInh, # stimulusThresholdInh
max(spTest._overlaps) / 1000, # addToWinners
)
# update _dutyCycleAfterInh
spTest._onCells.fill(0)
onCellIndices = spTest._onCellIndices[0:numOn]
spTest._onCells[onCellIndices] = 1
denseOn = spTest._onCells
spTest._dutyCycleAfterInh = (
((spTest.dutyCyclePeriod - 1) * spTest._dutyCycleAfterInh +
denseOn) / spTest.dutyCyclePeriod)
# learning step
spTest._adaptSynapses(onCellIndices, [], input_)
# update boostFactor
spTest._updateBoostFactors()
boostFactors = spTest._firingBoostFactors
# update dutyCycle and boost
if ((spTest._iterNum + 1) % 50) == 0:
spTest._updateInhibitionObj()
spTest._updateMinDutyCycles(spTest._dutyCycleBeforeInh,
spTest.minPctDutyCycleBeforeInh,
spTest._minDutyCycleBeforeInh)
spTest._updateMinDutyCycles(spTest._dutyCycleAfterInh,
spTest.minPctDutyCycleAfterInh,
spTest._minDutyCycleAfterInh)
# test numOn and spTest.numActiveColumnsPerInhArea
self.assertEqual(
numOn, spTest.numActiveColumnsPerInhArea,
"Error at input %s, actual numOn are: %i, "
"numActivePerInhAre is: %s" %
(i, numOn, numActiveColumnsPerInhArea))
# Spatial Pooler Parameters
var_encoders = {Encoder('value', ['RDSE'])}
# Encoder('_index', ['TODE'])}
encoder_width = 0
for x in var_encoders:
for y in x.encoders:
exec("s = " + y + ".getWidth()")
encoder_width += s
SP = SpatialPooler(
inputDimensions=encoder_width,
columnDimensions=pooler_out,
potentialPct=0.8,
globalInhibition=True,
numActiveColumnsPerInhArea=pooler_out // 50, # Gets 2% of the total area
boostStrength=1.0,
wrapAround=False)
TM = TemporalMemory(columnDimensions=(pooler_out, ), cellsPerColumn=cell_col)
# Train Spatial Pooler
start = time.time()
active_columns = np.zeros(pooler_out)
print("Spatial pooler learning")
for x in range(len(Data)):
encoder = multiencode(var_encoders, Data, x)
# e_val = RDSE.encode(Data['value'][x])
def runHotgym(numRecords):
with open(_PARAMS_PATH, "r") as f:
modelParams = yaml.safe_load(f)["modelParams"]
enParams = modelParams["sensorParams"]["encoders"]
spParams = modelParams["spParams"]
tmParams = modelParams["tmParams"]
timeOfDayEncoder = DateEncoder(
timeOfDay=enParams["timestamp_timeOfDay"]["timeOfDay"])
weekendEncoder = DateEncoder(
weekend=enParams["timestamp_weekend"]["weekend"])
scalarEncoder = RandomDistributedScalarEncoder(
enParams["consumption"]["resolution"])
encodingWidth = (timeOfDayEncoder.getWidth()
+ weekendEncoder.getWidth()
+ scalarEncoder.getWidth())
sp = SpatialPooler(
# How large the input encoding will be.
inputDimensions=(encodingWidth),
# How many mini-columns will be in the Spatial Pooler.
columnDimensions=(spParams["columnCount"]),
# What percent of the columns"s receptive field is available for potential
# synapses?
potentialPct=spParams["potentialPct"],
# This means that the input space has no topology.
globalInhibition=spParams["globalInhibition"],
localAreaDensity=spParams["localAreaDensity"],
# Roughly 2%, giving that there is only one inhibition area because we have
# turned on globalInhibition (40 / 2048 = 0.0195)
numActiveColumnsPerInhArea=spParams["numActiveColumnsPerInhArea"],
# How quickly synapses grow and degrade.
synPermInactiveDec=spParams["synPermInactiveDec"],
synPermActiveInc=spParams["synPermActiveInc"],
synPermConnected=spParams["synPermConnected"],
# boostStrength controls the strength of boosting. Boosting encourages
# efficient usage of SP columns.
boostStrength=spParams["boostStrength"],
# Random number generator seed.
seed=spParams["seed"],
# TODO: is this useful?
# Determines if inputs at the beginning and end of an input dimension should
# be considered neighbors when mapping columns to inputs.
wrapAround=False
)
tm = TemporalMemory(
# Must be the same dimensions as the SP
columnDimensions=(tmParams["columnCount"],),
# How many cells in each mini-column.
cellsPerColumn=tmParams["cellsPerColumn"],
# A segment is active if it has >= activationThreshold connected synapses
# that are active due to infActiveState
activationThreshold=tmParams["activationThreshold"],
initialPermanence=tmParams["initialPerm"],
# TODO: This comes from the SP params, is this normal
connectedPermanence=spParams["synPermConnected"],
# Minimum number of active synapses for a segment to be considered during
# search for the best-matching segments.
minThreshold=tmParams["minThreshold"],
# The max number of synapses added to a segment during learning
maxNewSynapseCount=tmParams["newSynapseCount"],
permanenceIncrement=tmParams["permanenceInc"],
permanenceDecrement=tmParams["permanenceDec"],
predictedSegmentDecrement=0.0,
maxSegmentsPerCell=tmParams["maxSegmentsPerCell"],
maxSynapsesPerSegment=tmParams["maxSynapsesPerSegment"],
seed=tmParams["seed"]
)
classifier = SDRClassifierFactory.create()
results = []
with open(_INPUT_FILE_PATH, "r") as fin:
reader = csv.reader(fin)
headers = reader.next()
reader.next()
reader.next()
for count, record in enumerate(reader):
if count >= numRecords: break
# Convert data string into Python date object.
dateString = datetime.datetime.strptime(record[0], "%m/%d/%y %H:%M")
# Convert data value string into float.
consumption = float(record[1])
# To encode, we need to provide zero-filled numpy arrays for the encoders
# to populate.
timeOfDayBits = numpy.zeros(timeOfDayEncoder.getWidth())
weekendBits = numpy.zeros(weekendEncoder.getWidth())
consumptionBits = numpy.zeros(scalarEncoder.getWidth())
# Now we call the encoders create bit representations for each value.
timeOfDayEncoder.encodeIntoArray(dateString, timeOfDayBits)
weekendEncoder.encodeIntoArray(dateString, weekendBits)
scalarEncoder.encodeIntoArray(consumption, consumptionBits)
# Concatenate all these encodings into one large encoding for Spatial
# Pooling.
encoding = numpy.concatenate(
[timeOfDayBits, weekendBits, consumptionBits]
)
# Create an array to represent active columns, all initially zero. This
# will be populated by the compute method below. It must have the same
# dimensions as the Spatial Pooler.
activeColumns = numpy.zeros(spParams["columnCount"])
# Execute Spatial Pooling algorithm over input space.
sp.compute(encoding, True, activeColumns)
activeColumnIndices = numpy.nonzero(activeColumns)[0]
# Execute Temporal Memory algorithm over active mini-columns.
tm.compute(activeColumnIndices, learn=True)
activeCells = tm.getActiveCells()
# Get the bucket info for this input value for classification.
bucketIdx = scalarEncoder.getBucketIndices(consumption)[0]
# Run classifier to translate active cells back to scalar value.
classifierResult = classifier.compute(
recordNum=count,
patternNZ=activeCells,
classification={
"bucketIdx": bucketIdx,
"actValue": consumption
},
learn=True,
infer=True
)
# Print the best prediction for 1 step out.
oneStepConfidence, oneStep = sorted(
zip(classifierResult[1], classifierResult["actualValues"]),
reverse=True
)[0]
print("1-step: {:16} ({:4.4}%)".format(oneStep, oneStepConfidence * 100))
results.append([oneStep, oneStepConfidence * 100, None, None])
return results
def __init__(self, b1, b2, *args, **kw_args): self.b1 = b1 self.b2 = b2 SpatialPooler.__init__(self, *args, **kw_args) self.reset()
def _runLearnInference(self,
n=30,
w=15,
columnDimensions=2048,
numActiveColumnsPerInhArea=40,
spSeed=1951,
spVerbosity=0,
numTrainingRecords=100,
seed=42):
# Instantiate two identical spatial pooler. One will be used only for
# learning. The other will be trained with identical records, but with
# random inference calls thrown in
spLearnOnly = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, n),
potentialRadius=n/2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
seed=spSeed,
synPermInactiveDec=0.01,
synPermActiveInc=0.2,
synPermConnected=0.11,)
spLearnInfer = SpatialPooler(
columnDimensions=(columnDimensions, 1),
inputDimensions=(1, n),
potentialRadius=n/2,
numActiveColumnsPerInhArea=numActiveColumnsPerInhArea,
spVerbosity=spVerbosity,
seed=spSeed,
synPermInactiveDec=0.01,
synPermActiveInc=0.2,
synPermConnected=0.11,)
random.seed(seed)
np.random.seed(seed)
# Build up training set with numTrainingRecords patterns
inputs = [] # holds post-encoded input patterns
for i in xrange(numTrainingRecords):
inputVector = np.zeros(n, dtype=realDType)
inputVector [random.sample(xrange(n), w)] = 1
inputs.append(inputVector)
# Train each SP with identical inputs
startTime = time.time()
random.seed(seed)
np.random.seed(seed)
for i in xrange(numTrainingRecords):
if spVerbosity > 0:
print "Input #%d" % i
# TODO: See https://github.com/numenta/nupic/issues/2072
encodedInput = inputs[i]
decodedOutput = np.zeros(columnDimensions)
spLearnOnly.compute(encodedInput, learn=True, activeArray=decodedOutput)
random.seed(seed)
np.random.seed(seed)
for i in xrange(numTrainingRecords):
if spVerbosity > 0:
print "Input #%d" % i
# TODO: See https://github.com/numenta/nupic/issues/2072
encodedInput = inputs[i]
decodedOutput = np.zeros(columnDimensions)
spLearnInfer.compute(encodedInput, learn=True, activeArray=decodedOutput)
print "\nElapsed time: %.2f seconds\n" % (time.time() - startTime)
# Test that both SP"s are identical by checking learning stats
# A more in depth test would check all the coincidences, duty cycles, etc.
# ala tpDiff
# Edit: spDiff has been written as an in depth tester of the spatial pooler
learnOnlyStats = spLearnOnly.getLearningStats()
learnInferStats = spLearnInfer.getLearningStats()
success = True
# Check that the two spatial poolers are equivalent after the same training.
success = success and spDiff(spLearnInfer, spLearnOnly)
self.assertTrue(success)
# Make sure that the pickled and loaded SPs are equivalent.
spPickle = pickle.dumps(spLearnOnly, protocol=0)
spLearnOnlyLoaded = pickle.loads(spPickle)
success = success and spDiff(spLearnOnly, spLearnOnlyLoaded)
self.assertTrue(success)
for k in learnOnlyStats.keys():
if learnOnlyStats[k] != learnInferStats[k]:
success = False
print "Stat", k, "is different:", learnOnlyStats[k], learnInferStats[k]
self.assertTrue(success)
if success:
print "Test succeeded"
def toString(self):
print("sentenceNum: " + self.id)
print("startIdx " + self.startIdx)
print("startIdx " + self.endIdx)
print("anomalyScore " + self.anomalyScore)
# parameters for the spatial pooler and temporal memory networks, only tm_only is used
sp_layer1 = SpatialPooler(inputDimensions=(128, 128),
columnDimensions=(64, 64),
potentialPct=0.1,
potentialRadius=5,
globalInhibition=False,
localAreaDensity=0.1,
numActiveColumnsPerInhArea=3,
synPermInactiveDec=0.5,
synPermActiveInc=0.02,
synPermConnected=0.90,
boostStrength=0.0,
wrapAround=False)
tm_only = TemporalMemory(
inputDimensions=(4096, ),
columnDimensions=(4096, ),
cellsPerColumn=5,
newSynapseCount=15,
activationThreshold=15,
initialPermanence=0.7,
connectedPermanence=0.8,
minThreshold=8,
uintType = "uint32" inputDimensions = (1000,1) columnDimensions = (2048,1) inputSize = np.array(inputDimensions).prod() columnNumber = np.array(columnDimensions).prod() inputArray = np.zeros(inputSize, dtype=uintType) for i in range(inputSize): inputArray[i] = random.randrange(2) activeCols = np.zeros(columnNumber, dtype=uintType) sp = SP(inputDimensions, columnDimensions, potentialRadius = int(0.5*inputSize), numActiveColumnsPerInhArea = int(0.02*columnNumber), globalInhibition = True, seed = 1, synPermActiveInc = 0.01, synPermInactiveDec = 0.008 ) # Part 1: # ------- # A column connects to a subset of the input vector (specified # by both the potentialRadius and potentialPct). The overlap score # for a column is the number of connections to the input that become # active when presented with a vector. When learning is 'on' in the SP, # the active connections are reinforced, whereas those inactive are # depressed (according to parameters synPermActiveInc and synPermInactiveDec. # In order for the SP to create a sparse representation of the input, it # will select a small percentage (usually 2%) of its most active columns,
# minval=0,
# maxval=100)
#consumeEncoder = AdaptiveScalarEncoder(
# n=400,
# w=21)
consumeEncoder = SimHashDistributedScalarEncoder(n=400, w=21, resolution=0.25)
encodingWidth = (timeOfDayEncoder.getWidth() + weekendEncoder.getWidth() +
consumeEncoder.getWidth())
classifier = SDRClassifierFactory.create()
sp = SpatialPooler(inputDimensions=(encodingWidth, ),
columnDimensions=(COL_WIDTH),
potentialPct=0.85,
potentialRadius=encodingWidth,
globalInhibition=True,
localAreaDensity=-1.0,
numActiveColumnsPerInhArea=40,
synPermInactiveDec=0.005,
synPermActiveInc=0.04,
synPermConnected=0.1,
boostStrength=3.0,
seed=1956,
wrapAround=False)
tm = TemporalMemory(columnDimensions=(COL_WIDTH, ),
cellsPerColumn=32,
activationThreshold=16,
initialPermanence=0.21,
connectedPermanence=0.5,
minThreshold=12,
maxNewSynapseCount=20,
permanenceIncrement=0.1,
permanenceDecrement=0.1,
def __init__(self, columnCount, InputEncoderParams, toL4ConnectorParamsI,
toL4ConnectorParamsII, toL5ConnectorParams,
toD1ConnectorParams, toD2ConnectorParams, L4Params, L5Params,
k, D1Params, D2Params):
self.columnCount = columnCount
self.toL4ConnectorParamsI = toL4ConnectorParamsI
self.toL4ConnectorParamsII = toL4ConnectorParamsII
self.toL5ConnectorParams = toL5ConnectorParams
self.toD1ConnectorParams = toD1ConnectorParams
self.toD2ConnectorParams = toD2ConnectorParams
self.L4Params = L4Params
self.L5Params = L5Params
self.k = k
self.D1Params = D1Params
self.D2Params = D2Params
self.learning = False
#encoder
from nupic.encoders import MultiEncoder
self.InputEncoder = MultiEncoder()
self.InputEncoder.addMultipleEncoders(InputEncoderParams)
print "Encoder Online"
#spatialPoolers
from nupic.algorithms.spatial_pooler import SpatialPooler
self.toL4ConnectorI = SpatialPooler(
inputDimensions=(toL4ConnectorParamsI["inputDimensions"], ),
columnDimensions=(columnCount, ),
potentialPct=toL4ConnectorParamsI["potentialPct"],
globalInhibition=toL4ConnectorParamsI["globalInhibition"],
localAreaDensity=toL4ConnectorParamsI["localAreaDensity"],
numActiveColumnsPerInhArea=toL4ConnectorParamsI[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL4ConnectorParamsI["synPermInactiveDec"],
synPermActiveInc=toL4ConnectorParamsI["synPermActiveInc"],
synPermConnected=toL4ConnectorParamsI["synPermConnected"],
boostStrength=toL4ConnectorParamsI["boostStrength"],
seed=toL4ConnectorParamsI["seed"],
wrapAround=toL4ConnectorParamsI["wrapAround"]) #this part sucks
self.toL4ConnectorII = SpatialPooler(
inputDimensions=(columnCount * 3, ),
columnDimensions=(columnCount, ),
potentialPct=toL4ConnectorParamsII["potentialPct"],
globalInhibition=toL4ConnectorParamsII["globalInhibition"],
localAreaDensity=toL4ConnectorParamsII["localAreaDensity"],
numActiveColumnsPerInhArea=toL4ConnectorParamsII[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL4ConnectorParamsII["synPermInactiveDec"],
synPermActiveInc=toL4ConnectorParamsII["synPermActiveInc"],
synPermConnected=toL4ConnectorParamsII["synPermConnected"],
boostStrength=toL4ConnectorParamsII["boostStrength"],
seed=toL4ConnectorParamsII["seed"],
wrapAround=toL4ConnectorParamsII["wrapAround"])
print "toL4Connector Online"
self.toL5Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toL5ConnectorParams["potentialPct"],
globalInhibition=toL5ConnectorParams["globalInhibition"],
localAreaDensity=toL5ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toL5ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toL5ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toL5ConnectorParams["synPermActiveInc"],
synPermConnected=toL5ConnectorParams["synPermConnected"],
boostStrength=toL5ConnectorParams["boostStrength"],
seed=toL5ConnectorParams["seed"],
wrapAround=toL5ConnectorParams["wrapAround"])
print "toL5Connector Online"
self.toD1Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toD1ConnectorParams["potentialPct"],
globalInhibition=toD1ConnectorParams["globalInhibition"],
localAreaDensity=toD1ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toD1ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toD1ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toD1ConnectorParams["synPermActiveInc"],
synPermConnected=toD1ConnectorParams["synPermConnected"],
boostStrength=toD1ConnectorParams["boostStrength"],
seed=toD1ConnectorParams["seed"],
wrapAround=toD1ConnectorParams["wrapAround"])
print "toD1Connector Online"
self.toD2Connector = SpatialPooler(
inputDimensions=(columnCount, ),
columnDimensions=(columnCount, ),
potentialPct=toD2ConnectorParams["potentialPct"],
globalInhibition=toD2ConnectorParams["globalInhibition"],
localAreaDensity=toD2ConnectorParams["localAreaDensity"],
numActiveColumnsPerInhArea=toD2ConnectorParams[
"numActiveColumnsPerInhArea"],
synPermInactiveDec=toD2ConnectorParams["synPermInactiveDec"],
synPermActiveInc=toD2ConnectorParams["synPermActiveInc"],
synPermConnected=toD2ConnectorParams["synPermConnected"],
boostStrength=toD2ConnectorParams["boostStrength"],
seed=toD2ConnectorParams["seed"],
wrapAround=toD2ConnectorParams["wrapAround"])
print "toD2Connector Online"
#HTM Layers
from nupic.algorithms.temporal_memory import TemporalMemory
self.L4ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.L4 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
)
print "L4 Online"
self.L5ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.L5 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
)
print "L5 Online"
self.D1ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.D1 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
initialPermanence=0.21,
connectedPermanence=0.5,
)
print "D1 Online"
self.D2ActiveColumns = numpy.zeros(self.columnCount, dtype=int)
self.D2 = TemporalMemory(
columnDimensions=(columnCount, ),
seed=42,
initialPermanence=0.21,
connectedPermanence=0.5,
)
print "D2 Online"
def testSP():
""" Run a SP test
"""
elemSize = 400
numSet = 42
addNear = True
numRecords = 2
wantPlot = True
poolPct = 0.5
itr = 1
doLearn = True
while numRecords < 3:
# Setup a SP
sp = SpatialPooler(
columnDimensions=(2048, 1),
inputDimensions=(1, elemSize),
potentialRadius=elemSize/2,
numActiveColumnsPerInhArea=40,
spVerbosity=0,
stimulusThreshold=0,
seed=1,
potentialPct=poolPct,
globalInhibition=True
)
# Generate inputs using rand()
inputs = generateRandomInput(numRecords, elemSize, numSet)
if addNear:
# Append similar entries (distance of 1)
appendInputWithNSimilarValues(inputs, 42)
inputSize = len(inputs)
print 'Num random records = %d, inputs to process %d' % (numRecords, inputSize)
# Run a number of iterations, with learning on or off,
# retrieve results from the last iteration only
outputs = np.zeros((inputSize,2048))
numIter = 1
if doLearn:
numIter = itr
for iter in xrange(numIter):
for i in xrange(inputSize):
time.sleep(0.001)
if iter == numIter - 1:
# TODO: See https://github.com/numenta/nupic/issues/2072
sp.compute(inputs[i], learn=doLearn, activeArray=outputs[i])
#print outputs[i].sum(), outputs[i]
else:
# TODO: See https://github.com/numenta/nupic/issues/2072
output = np.zeros(2048)
sp.compute(inputs[i], learn=doLearn, activeArray=output)
# Build a plot from the generated input and output and display it
distribMatrix = generatePlot(outputs, inputs)
# If we don't want a plot, just continue
if wantPlot:
plt.imshow(distribMatrix, origin='lower', interpolation = "nearest")
plt.ylabel('SP (2048/40) distance in %')
plt.xlabel('Input (400/42) distance in %')
title = 'SP distribution'
if doLearn:
title += ', leaning ON'
else:
title += ', learning OFF'
title += ', inputs = %d' % len(inputs)
title += ', iterations = %d' % numIter
title += ', poolPct =%f' % poolPct
plt.suptitle(title, fontsize=12)
plt.show()
#plt.savefig(os.path.join('~/Desktop/ExperimentResults/videos5', '%s' % numRecords))
#plt.clf()
numRecords += 1
return
class DendriteDetector(AnomalyDetector):
def initialize(self):
# Keep track of value range for spatial anomaly detection.
self.minVal = None
self.maxVal = None
# Time of day encoder
self.timeOfDayEncoder = DateEncoder(timeOfDay=(21, 9.49),
name='time_enc')
# RDSE encoder for the time series value.
minResolution = 0.001
rangePadding = abs(self.inputMax - self.inputMin) * 0.2
minVal = self.inputMin - rangePadding
maxVal = self.inputMax + rangePadding
numBuckets = 130
resolution = max(minResolution, (maxVal - minVal) / numBuckets)
self.value_enc = RandomDistributedScalarEncoder(resolution=resolution,
name='value_rdse')
# Spatial Pooler.
encodingWidth = self.timeOfDayEncoder.getWidth(
) + self.value_enc.getWidth()
self.sp = SpatialPooler(
inputDimensions=(encodingWidth, ),
columnDimensions=(2048, ),
potentialPct=0.8,
potentialRadius=encodingWidth,
globalInhibition=1,
numActiveColumnsPerInhArea=40,
synPermInactiveDec=0.0005,
synPermActiveInc=0.003,
synPermConnected=0.2,
boostStrength=0.0,
seed=1956,
wrapAround=True,
)
self.tm = TemporalMemory(
columnDimensions=(2048, ),
cellsPerColumn=32,
activationThreshold=20,
initialPermanence=.5, # Increased to connectedPermanence.
connectedPermanence=.5,
minThreshold=13,
maxNewSynapseCount=31,
permanenceIncrement=0.04,
permanenceDecrement=0.008,
predictedSegmentDecrement=0.001,
maxSegmentsPerCell=128,
maxSynapsesPerSegment=
128, # Changed meaning. Also see connections.topology[2]
seed=1993,
)
# Initialize the anomaly likelihood object
numentaLearningPeriod = int(math.floor(self.probationaryPeriod / 2.0))
self.anomalyLikelihood = anomaly_likelihood.AnomalyLikelihood(
learningPeriod=numentaLearningPeriod,
estimationSamples=self.probationaryPeriod - numentaLearningPeriod,
reestimationPeriod=100,
)
self.age = 0
def getAdditionalHeaders(self):
"""Returns a list of strings."""
return ["raw_score"]
def handleRecord(self, inputData):
"""
Argument inputData is {"value": instantaneous_value, "timestamp": pandas.Timestamp}
Returns a tuple (anomalyScore, rawScore).
Internally to NuPIC "anomalyScore" corresponds to "likelihood_score"
and "rawScore" corresponds to "anomaly_score". Sorry about that.
"""
# Check for spatial anomalies and update min/max values.
value = inputData["value"]
spatialAnomaly = False
if self.minVal != self.maxVal:
tolerance = (self.maxVal - self.minVal) * SPATIAL_TOLERANCE
maxExpected = self.maxVal + tolerance
minExpected = self.minVal - tolerance
if value > maxExpected or value < minExpected:
spatialAnomaly = True
if self.maxVal is None or value > self.maxVal:
self.maxVal = value
if self.minVal is None or value < self.minVal:
self.minVal = value
# Run the HTM stack. First Encoders.
timestamp = inputData["timestamp"]
timeOfDayBits = np.zeros(self.timeOfDayEncoder.getWidth())
self.timeOfDayEncoder.encodeIntoArray(timestamp, timeOfDayBits)
valueBits = np.zeros(self.value_enc.getWidth())
self.value_enc.encodeIntoArray(value, valueBits)
encoding = np.concatenate([timeOfDayBits, valueBits])
# Spatial Pooler.
activeColumns = np.zeros(self.sp.getNumColumns())
self.sp.compute(encoding, True, activeColumns)
activeColumnIndices = np.nonzero(activeColumns)[0]
# Temporal Memory and Anomaly.
predictions = self.tm.getPredictiveCells()
predictedColumns = list(self.tm.mapCellsToColumns(predictions).keys())
self.tm.compute(activeColumnIndices, learn=True)
activeCells = self.tm.getActiveCells()
rawScore = anomaly.computeRawAnomalyScore(activeColumnIndices,
predictedColumns)
# Compute log(anomaly likelihood)
anomalyScore = self.anomalyLikelihood.anomalyProbability(
inputData["value"], rawScore, inputData["timestamp"])
finalScore = logScore = self.anomalyLikelihood.computeLogLikelihood(
anomalyScore)
if spatialAnomaly:
finalScore = 1.0
if False:
# Plot correlation of excitement versus compartmentalization.
if self.age == 0:
print("Correlation Plots ENABLED.")
if False:
start_age = 1000
end_age = 1800
else:
start_age = 4000
end_age = 7260
if self.age == start_age:
import correlation
import random
self.cor_samplers = []
sampled_cells = []
while len(self.cor_samplers) < 20:
n = random.choice(xrange(self.tm.numberOfCells()))
if n in sampled_cells:
continue
else:
sampled_cells.append(n)
neuron = self.tm.connections.dataForCell(n)
if neuron._roots:
c = correlation.CorrelationSampler(neuron._roots[0])
c.random_sample_points(100)
self.cor_samplers.append(c)
print("Created %d Correlation Samplers" %
len(self.cor_samplers))
if self.age >= start_age:
for smplr in self.cor_samplers:
smplr.sample()
if self.age == end_age:
import matplotlib.pyplot as plt
for idx, smplr in enumerate(self.cor_samplers):
if smplr.num_samples == 0:
print("No samples, plot not shown.")
continue
plt.figure("Sample %d" % idx)
smplr.plot(period=64) # Different value!
plt.show()
if False:
# Plot excitement of a typical detection on a dendrite.
if self.age == 7265:
#if self.age == 1800:
import matplotlib.pyplot as plt
import random
from connections import SYN_CONNECTED_ACTIVE
sampled_cells = set()
for figure_num in xrange(40):
plt.figure("(%d)" % figure_num)
# Find an active cell to view.
cell = None
for attempt in range(100):
event = random.choice(self.tm.activeEvents)
cell = event.cell # This is an integer.
if cell is not None and cell not in sampled_cells:
break
else:
break
sampled_cells.add(cell)
cell = self.tm.connections.dataForCell(cell)
# Organize the data.
EPSPs = []
excitement = []
distance_to_root = 0
segment_offsets = {}
branch = cell._roots[0]
while True:
segment_offsets[branch] = distance_to_root
distance_to_root += len(branch._synapses)
excitement.extend(branch.excitement)
for syn in branch._synapses:
if syn is None:
EPSPs.append(0)
else:
EPSPs.append(syn.state == SYN_CONNECTED_ACTIVE)
if branch.children:
branch = random.choice(branch.children)
else:
break
plt.plot(
np.arange(distance_to_root),
EPSPs,
'r',
np.arange(distance_to_root),
excitement,
'b',
)
plt.title(
"Dendrite Activation\n Horizontal line is activation threshold, Vertical lines are segment bifurcations"
)
plt.xlabel("Distance along Dendrite", )
plt.ylabel("EPSPs are Red, Excitement is Blue")
# Show lines where the excitement crosses thresholds.
plt.axhline(20, color='k') # Hard coded parameter value.
for offset in segment_offsets.values():
if offset != 0:
plt.axvline(offset, color='k')
print("\nShowing %d excitement plots." % len(sampled_cells))
plt.show()
self.age += 1
return (finalScore, rawScore)
def testSPNew():
""" New version of the test"""
elemSize = 400
numSet = 42
addNear = True
numRecords = 1000
wantPlot = False
poolPct = 0.5
itr = 5
pattern = [60, 1000]
doLearn = True
start = 1
learnIter = 0
noLearnIter = 0
numLearns = 0
numTests = 0
numIter = 1
numGroups = 1000
PLOT_PRECISION = 100.0
distribMatrix = np.zeros((PLOT_PRECISION+1,PLOT_PRECISION+1))
inputs = generateRandomInput(numGroups, elemSize, numSet)
# Setup a SP
sp = SpatialPooler(
columnDimensions=(2048, 1),
inputDimensions=(1, elemSize),
potentialRadius=elemSize/2,
numActiveColumnsPerInhArea=40,
spVerbosity=0,
stimulusThreshold=0,
synPermConnected=0.12,
seed=1,
potentialPct=poolPct,
globalInhibition=True
)
cleanPlot = False
for i in xrange(numRecords):
input1 = getRandomWithMods(inputs, 4)
if i % 2 == 0:
input2 = getRandomWithMods(inputs, 4)
else:
input2 = input1.copy()
input2 = modifyBits(input2, 21)
inDist = (abs(input1-input2) > 0.1)
intInDist = int(inDist.sum()/2+0.1)
#print intInDist
if start == 0:
doLearn = True
learnIter += 1
if learnIter == pattern[start]:
numLearns += 1
start = 1
noLearnIter = 0
elif start == 1:
doLearn = False
noLearnIter += 1
if noLearnIter == pattern[start]:
numTests += 1
start = 0
learnIter = 0
cleanPlot = True
# TODO: See https://github.com/numenta/nupic/issues/2072
sp.compute(input1, learn=doLearn, activeArray=output1)
sp.compute(input2, learn=doLearn, activeArray=output2)
time.sleep(0.001)
outDist = (abs(output1-output2) > 0.1)
intOutDist = int(outDist.sum()/2+0.1)
if not doLearn and intOutDist < 2 and intInDist > 10:
"""
sp.spVerbosity = 10
# TODO: See https://github.com/numenta/nupic/issues/2072
sp.compute(input1, learn=doLearn, activeArray=output1)
sp.compute(input2, learn=doLearn, activeArray=output2)
sp.spVerbosity = 0
print 'Elements has very small SP distance: %d' % intOutDist
print output1.nonzero()
print output2.nonzero()
print sp._firingBoostFactors[output1.nonzero()[0]]
print sp._synPermBoostFactors[output1.nonzero()[0]]
print 'Input elements distance is %d' % intInDist
print input1.nonzero()
print input2.nonzero()
sys.stdin.readline()
"""
if not doLearn:
x = int(PLOT_PRECISION*intOutDist/40.0)
y = int(PLOT_PRECISION*intInDist/42.0)
if distribMatrix[x, y] < 0.1:
distribMatrix[x, y] = 3
else:
if distribMatrix[x, y] < 10:
distribMatrix[x, y] += 1
#print i
# If we don't want a plot, just continue
if wantPlot and cleanPlot:
plt.imshow(distribMatrix, origin='lower', interpolation = "nearest")
plt.ylabel('SP (2048/40) distance in %')
plt.xlabel('Input (400/42) distance in %')
title = 'SP distribution'
#if doLearn:
# title += ', leaning ON'
#else:
# title += ', learning OFF'
title += ', learn sets = %d' % numLearns
title += ', test sets = %d' % numTests
title += ', iter = %d' % numIter
title += ', groups = %d' % numGroups
title += ', Pct =%f' % poolPct
plt.suptitle(title, fontsize=12)
#plt.show()
plt.savefig(os.path.join('~/Desktop/ExperimentResults/videosNew', '%s' % i))
plt.clf()
distribMatrix = np.zeros((PLOT_PRECISION+1,PLOT_PRECISION+1))
cleanPlot = False
if __name__ == "__main__":
# Get training images and convert them to vectors.
trainingImages, trainingTags = dataset_readers.getImagesAndTags(
trainingDataset)
trainingVectors = encoder.imagesToVectors(trainingImages)
# Instantiate the python spatial pooler
sp = SpatialPooler(
inputDimensions=32**2, # Size of image patch
columnDimensions=16, # Number of potential features
potentialRadius=10000, # Ensures 100% potential pool
potentialPct=1, # Neurons can connect to 100% of input
globalInhibition=True,
localAreaDensity=-1, # Using numActiveColumnsPerInhArea
#localAreaDensity=0.02, # one percent of columns active at a time
#numActiveColumnsPerInhArea=-1, # Using percentage instead
numActiveColumnsPerInhArea=1, # Only one feature active at a time
# All input activity can contribute to feature output
stimulusThreshold=0,
synPermInactiveDec=0.3,
synPermActiveInc=0.3,
synPermConnected=0.3, # Connected threshold
boostStrength=2,
seed=1956, # The seed that Grok uses
spVerbosity=1)
# Instantiate the spatial pooler test bench.
tb = VisionTestBench(sp)
# Instantiate the classifier
clf = exactMatch()
def testSPFile():
""" Run test on the data file - the file has records previously encoded.
"""
spSize = 2048
spSet = 40
poolPct = 0.5
pattern = [50, 1000]
doLearn = True
PLOT_PRECISION = 100.0
distribMatrix = np.zeros((PLOT_PRECISION+1,PLOT_PRECISION+1))
inputs = []
#file = open('~/Desktop/ExperimentResults/sampleArtificial.csv', 'rb')
#elemSize = 400
#numSet = 42
#file = open('~/Desktop/ExperimentResults/sampleDataBasilOneField.csv', 'rb')
#elemSize = 499
#numSet = 7
outdir = '~/Desktop/ExperimentResults/Basil100x21'
inputFile = outdir+'.csv'
file = open(inputFile, 'rb')
elemSize = 100
numSet = 21
reader = csv.reader(file)
for row in reader:
input = np.array(map(float, row), dtype=realDType)
if len(input.nonzero()[0]) != numSet:
continue
inputs.append(input.copy())
file.close()
# Setup a SP
sp = SpatialPooler(
columnDimensions=(spSize, 1),
inputDimensions=(1, elemSize),
potentialRadius=elemSize/2,
numActiveColumnsPerInhArea=spSet,
spVerbosity=0,
stimulusThreshold=0,
synPermConnected=0.10,
seed=1,
potentialPct=poolPct,
globalInhibition=True
)
cleanPlot = False
doLearn = False
print 'Finished reading file, inputs/outputs to process =', len(inputs)
size = len(inputs)
for iter in xrange(100):
print 'Iteration', iter
# Learn
if iter != 0:
for learnRecs in xrange(pattern[0]):
# TODO: See https://github.com/numenta/nupic/issues/2072
ind = np.random.random_integers(0, size-1, 1)[0]
sp.compute(inputs[ind], learn=True, activeArray=outputs[ind])
# Test
for _ in xrange(pattern[1]):
rand1 = np.random.random_integers(0, size-1, 1)[0]
rand2 = np.random.random_integers(0, size-1, 1)[0]
sp.compute(inputs[rand1], learn=False, activeArray=output1)
sp.compute(inputs[rand2], learn=False, activeArray=output2)
outDist = (abs(output1-output2) > 0.1)
intOutDist = int(outDist.sum()/2+0.1)
inDist = (abs(inputs[rand1]-inputs[rand2]) > 0.1)
intInDist = int(inDist.sum()/2+0.1)
if intInDist != numSet or intOutDist != spSet:
print rand1, rand2, '-', intInDist, intOutDist
x = int(PLOT_PRECISION*intOutDist/spSet)
y = int(PLOT_PRECISION*intInDist/numSet)
if distribMatrix[x, y] < 0.1:
distribMatrix[x, y] = 3
else:
if distribMatrix[x, y] < 10:
distribMatrix[x, y] += 1
if True:
plt.imshow(distribMatrix, origin='lower', interpolation = "nearest")
plt.ylabel('SP (%d/%d) distance in pct' % (spSize, spSet))
plt.xlabel('Input (%d/%d) distance in pct' % (elemSize, numSet))
title = 'SP distribution'
title += ', iter = %d' % iter
title += ', Pct =%f' % poolPct
plt.suptitle(title, fontsize=12)
#plt.savefig(os.path.join('~/Desktop/ExperimentResults/videosArtData', '%s' % iter))
plt.savefig(os.path.join(outdir, '%s' % iter))
plt.clf()
distribMatrix = np.zeros((PLOT_PRECISION+1,PLOT_PRECISION+1))
import numpy as np
from tqdm import tqdm
from nupic.encoders.scalar import ScalarEncoder
from nupic.algorithms.spatial_pooler import SpatialPooler
from nupic.algorithms.temporal_memory import TemporalMemory
from nupic.algorithms.sdr_classifier import SDRClassifier
N = 900
x = np.sin(np.arange(N) * 2 * np.pi / 30.0)
inputDimensions = (256, )
columnDimensions = (512, )
encoder = ScalarEncoder(21, -1.0, 1.0, n=inputDimensions[0])
sp = SpatialPooler(inputDimensions=inputDimensions,
columnDimensions=columnDimensions,
globalInhibition=True,
numActiveColumnsPerInhArea=21)
tm = TemporalMemory(columnDimensions=columnDimensions)
c = SDRClassifier(steps=[1], alpha=0.1, actValueAlpha=0.1, verbosity=0)
x_true = x[1:]
x_predict = np.zeros(len(x) - 1)
for i, xi in tqdm(enumerate(x[:-1])):
encoded = encoder.encode(xi)
bucketIdx = np.where(encoded > 0)[0][0]
spd = np.zeros(columnDimensions[0])
sp.compute(encoded, True, spd)
active_indices = np.where(spd > 0)[0]
tm.compute(active_indices)
# Pick a combination of parameter values
parameters.nextCombination()
#parameters.nextRandomCombination()
synPermConn = parameters.getValue("synPermConn")
synPermDec = synPermConn * parameters.getValue("synPermDecFrac")
synPermInc = synPermConn * parameters.getValue("synPermIncFrac")
# Instantiate our spatial pooler
sp = SpatialPooler(
inputDimensions=(32, 32), # Size of image patch
columnDimensions=(32, 32),
potentialRadius=10000, # Ensures 100% potential pool
potentialPct=0.8,
globalInhibition=True,
localAreaDensity=-1, # Using numActiveColumnsPerInhArea
numActiveColumnsPerInhArea=64,
# All input activity can contribute to feature output
stimulusThreshold=0,
synPermInactiveDec=synPermDec,
synPermActiveInc=synPermInc,
synPermConnected=synPermConn,
boostStrength=1.0,
seed=1956, # The seed that Grok uses
spVerbosity=1)
# Instantiate the spatial pooler test bench.
tb = VisionTestBench(sp)
# Instantiate the classifier
clf = KNNClassifier()
# Train the spatial pooler on trainingVectors.
def runHotgym(numRecords):
with open(_PARAMS_PATH, "r") as f:
modelParams = yaml.safe_load(f)["modelParams"]
enParams = modelParams["sensorParams"]["encoders"]
spParams = modelParams["spParams"]
tmParams = modelParams["tmParams"]
scalarEncoder = RandomDistributedScalarEncoder(
enParams["consumption"]["resolution"])
scalarEncoder2 = RandomDistributedScalarEncoder(
enParams["consumption2"]["resolution"])
encodingWidth = (scalarEncoder.getWidth() + scalarEncoder2.getWidth())
sp = SpatialPooler(
inputDimensions=(encodingWidth, ),
columnDimensions=(spParams["columnCount"], ),
potentialPct=spParams["potentialPct"],
potentialRadius=encodingWidth,
globalInhibition=spParams["globalInhibition"],
localAreaDensity=spParams["localAreaDensity"],
numActiveColumnsPerInhArea=spParams["numActiveColumnsPerInhArea"],
synPermInactiveDec=spParams["synPermInactiveDec"],
synPermActiveInc=spParams["synPermActiveInc"],
synPermConnected=spParams["synPermConnected"],
boostStrength=spParams["boostStrength"],
seed=spParams["seed"],
wrapAround=True)
tm = TemporalMemory(
columnDimensions=(tmParams["columnCount"], ),
cellsPerColumn=tmParams["cellsPerColumn"],
activationThreshold=tmParams["activationThreshold"],
initialPermanence=tmParams["initialPerm"],
connectedPermanence=spParams["synPermConnected"],
minThreshold=tmParams["minThreshold"],
maxNewSynapseCount=tmParams["newSynapseCount"],
permanenceIncrement=tmParams["permanenceInc"],
permanenceDecrement=tmParams["permanenceDec"],
predictedSegmentDecrement=0.0,
maxSegmentsPerCell=tmParams["maxSegmentsPerCell"],
maxSynapsesPerSegment=tmParams["maxSynapsesPerSegment"],
seed=tmParams["seed"])
classifier = SDRClassifierFactory.create()
results = []
with open(_INPUT_FILE_PATH, "r") as fin:
reader = csv.reader(fin)
headers = reader.next()
reader.next()
reader.next()
output = output_anomaly_generic_v1.NuPICFileOutput(_FILE_NAME)
for count, record in enumerate(reader):
if count >= numRecords: break
# Convert data string into Python date object.
# dateString = datetime.datetime.strptime(record[0], "%m/%d/%y %H:%M")
# Convert data value string into float.
prediction = float(record[1])
prediction2 = float(record[2])
# To encode, we need to provide zero-filled numpy arrays for the encoders
# to populate.
consumptionBits = numpy.zeros(scalarEncoder.getWidth())
consumptionBits2 = numpy.zeros(scalarEncoder2.getWidth())
# Now we call the encoders to create bit representations for each value.
scalarEncoder.encodeIntoArray(prediction, consumptionBits)
scalarEncoder2.encodeIntoArray(prediction2, consumptionBits2)
# Concatenate all these encodings into one large encoding for Spatial
# Pooling.
encoding = numpy.concatenate([consumptionBits, consumptionBits2])
# Create an array to represent active columns, all initially zero. This
# will be populated by the compute method below. It must have the same
# dimensions as the Spatial Pooler.
activeColumns = numpy.zeros(spParams["columnCount"])
# Execute Spatial Pooling algorithm over input space.
sp.compute(encoding, True, activeColumns)
activeColumnIndices = numpy.nonzero(activeColumns)[0]
# Execute Temporal Memory algorithm over active mini-columns.
tm.compute(activeColumnIndices, learn=True)
activeCells = tm.getActiveCells()
# Get the bucket info for this input value for classification.
bucketIdx = scalarEncoder.getBucketIndices(prediction)[0]
# Run classifier to translate active cells back to scalar value.
classifierResult = classifier.compute(recordNum=count,
patternNZ=activeCells,
classification={
"bucketIdx": bucketIdx,
"actValue": prediction
},
learn=True,
infer=True)
# Print the best prediction for 1 step out.
oneStepConfidence, oneStep = sorted(zip(
classifierResult[1], classifierResult["actualValues"]),
reverse=True)[0]
# print("1-step: {:16} ({:4.4}%)".format(oneStep, oneStepConfidence * 100))
# results.append([oneStep, oneStepConfidence * 100, None, None])
results.append(
[record[0], prediction, oneStep, oneStepConfidence * 100])
output.write(record[0], prediction, oneStep,
oneStepConfidence * 100)
output.close()
return results
EncodingWidth, SpatialPoolerWidth1 = 750, 600
sp = SpatialPooler(
# How large the input encoding will be.
inputDimensions=(EncodingWidth),
# How many mini-columns will be in the Spatial Pooler.
columnDimensions=(SpatialPoolerWidth1),
# What percent of the columns's receptive field is available for potential
# synapses?
potentialPct=0.85,
# This means that the input space has no topology.
globalInhibition=True,
localAreaDensity=-1.0,
# Roughly 2%, giving that there is only one inhibition area because we have
# turned on globalInhibition (40 / 2048 = 0.0195)
numActiveColumnsPerInhArea=30.0,
# How quickly synapses grow and degrade.
synPermInactiveDec=0.005,
synPermActiveInc=0.04,
synPermConnected=0.1,
# boostStrength controls the strength of boosting. Boosting encourages
# efficient usage of SP columns.
boostStrength=3.0,
# Random number generator seed.
seed=1956,
# Determines if inputs at the beginning and end of an input dimension should
# be considered neighbors when mapping columns to inputs.
wrapAround=False)
# Array which contains the output of the spatial pooler for layer 1
activeColumns = np.zeros(SpatialPoolerWidth1)
