def runOneExperiment(modelConfig,
inferenceArgs,
metricSpecs,
sourceSpec,
sinkSpec=None):
client = Client(modelConfig, inferenceArgs, metricSpecs, sourceSpec,
sinkSpec)
return client.run().metrics
def run(self):
self.initM(modelData)
while not self.kill_received:
jobaux = self.work_queue.get()
command = jobaux[0]
if command == 'predict':
self.index = self.index + 1
self.updateModelStats()
self.result_queue.put([
(self.Scores[m], self.predictionStreams[m][-1],
self.truth[self.index], m) for m in list(self.M.keys())
])
if command == 'getPredictionStreams':
self.result_queue.put(
dict([(m, self.predictionStreams[m][:-windowSize])
for m in list(self.predictionStreams.keys())]))
if command == 'delete':
delList = jobaux[1]
for d in delList:
if (d in self.M):
del self.M[d]
del self.Scores[d]
del self.predictionStreams[d]
print('deleted Model' + str(d) + " in process " +
str(self.iden))
print("number of models remaining in " + str(self.iden) +
": " + str(len(self.M)))
self.result_queue.put(self.iden)
if command == 'getAAEs':
self.result_queue.put([
(m, computeAAE(self.truth, self.predictionStreams[m],
r), self.getModelState(self.M[m]),
self.M[m]['modelDescription'])
for m in list(self.M.keys())
])
if command == 'addPSOVariants':
for t in jobaux[1]:
if (t[0] == self.iden):
name = t[2]
modelDescription = t[1][0]
x = t[1][1]
v = t[1][2]
self.M[name] = {}
self.M[name]['modelDescription'] = modelDescription
self.M[name]['client'] = Client(**modelDescription)
self.M[name]['alive'] = True
self.M[name]['start'] = 0
self.M[name]['end'] = None
self.M[name]['x'] = x
self.M[name]['v'] = v
self.Scores[name] = 10000
self.predictionStreams[name] = [
0,
]
print("added new model " + str(name) + " to process" +
str(self.iden))
def initM(self, modelDatList):
for modelData in modelDatList:
name = modelData[0]
self.M[name] = {}
self.M[name]['modelDescription'] = modelData[1]
self.M[name]['client'] = Client(**modelData[1])
alpha = modelData[1]['modelConfig']['modelParams']['clParams'][
'alpha']
n = 0
for encoder in modelData[1]['modelConfig']['modelParams'][
'sensorParams']['encoders']:
if encoder['name'] == predictedField:
n = encoder['n']
synPermInactiveDec = modelData[1]['modelConfig']['modelParams'][
'spParams']['synPermInactiveDec']
activationThreshold = modelData[1]['modelConfig']['modelParams'][
'tmParams']['activationThreshold']
pamLength = modelData[1]['modelConfig']['modelParams']['tmParams'][
'pamLength']
self.M[name]['x'] = np.array(
[alpha, n, synPermInactiveDec, activationThreshold, pamLength])
vAlpha = uniform(0.01, 0.15)
vN = randrange(30, 200, 5)
vSynPermInactiveDec = uniform(0.01, 0.15)
vActivationThreshold = randrange(12, 17, 1)
vPamLength = randrange(1, 6, 1)
self.M[name]['v'] = np.array([
vAlpha, vN, vSynPermInactiveDec, vActivationThreshold,
vPamLength
])
self.M[name]['alive'] = True
self.M[name]['start'] = 0
self.M[name]['end'] = None
self.Scores[name] = 10000
self.predictionStreams[name] = [
0,
]
def runOneExperiment(modelConfig, inferenceArgs, metricSpecs, sourceSpec,
sinkSpec=None):
client = Client(modelConfig, inferenceArgs, metricSpecs, sourceSpec, sinkSpec)
return client.run().metrics