def __init__(self, questionID, algorithm, classifier, parameters, modelParams, optimizer, predict = 0): self.questionID = questionID self.algorithm = algorithm self.classifier = classifier self.parameters = parameters self.modelParams = modelParams self.api = nemoApi() self.config = nemoConfig() self.optimizer = optimizer self.predict = predict self.prediction = None
def main():
global API, CONFIG, QUEUE, finishedRun, dataLoad
# Creating logger for logging to MAsterLog.log and console
# logger = createLogger()
# Load config file
CONFIG = nemoConfig('config/dev.json')
if CONFIG.CONFIG is None:
print 'Could not load config file on . . .'
return
# Start zerorpc server for remote control
t = threading.Thread(target=server)
t.start()
# Set semaphore
SEMAPHORE = threading.BoundedSemaphore(CONFIG.MAX_NUM_THREADS)
# Instantiate api for use by sub modules
API = nemoApi(CONFIG.HOST, CONFIG.PORT, CONFIG.USER, CONFIG.PASS, CONFIG.DB)
QUEUE = Queue.Queue()
# Run indefinitely
while True:
RESULTS = API.fetchQuestions(CONFIG.MAX_QUEUE_SIZE, CONFIG.QUEUED_STATUS)
for ROW in RESULTS:
QUEUE.put({
'id': ROW['ID'],
'makePrediction': ROW['MakePrediction'],
'Optimizer': ROW['Optimizer'],
'Classifier': ROW['Classifier']
})
if QUEUE.empty():
finishedRun = True
time.sleep(CONFIG.TIMEOUT)
else:
finishedRun = False
while not QUEUE.empty():
SEMAPHORE.acquire()
t = threading.Thread(target=worker, args=(QUEUE.get(), SEMAPHORE))
t.start()
QUEUE.join()
finishedRun = True
while dataLoad:
time.sleep(30)
print "Waiting on data load"
def main():
# Load config file
CONFIG = nemoConfig('config/dev.json')
# Instantiate api
API = nemoApi(CONFIG.HOST, CONFIG.PORT, CONFIG.USER, CONFIG.PASS, CONFIG.DB)
newWrapper = WekaWrapper(411, 'RandomForest', 'weka.classifiers.meta.CVParameterSelection', ["-W", "weka.classifiers.trees.RandomForest", "-P", "K 1 3 3"], [], '', predict=0)
newWrapper.run()
#masterData = newWrapper.retrieveData(336, 'all')
#learnerData = newWrapper.retrieveData(336, 'learner')
#testData = newWrapper.retrieveData(312, 'test')
#masterData.delete()
#patient = API.fetchPatientJSON(336)
#patientObj = newWrapper.buildPatientObject()
x = API.getDataQuery(411, 'ALL')
pdb.set_trace()
def run(id, optimizer, algorithm):
api = nemoApi()
config = nemoConfig()
# Fetch question info
info = api.fetchQuestionInfo(id)
# Declare vars to collect
parameters = []
options = None
tweak = None
tweakParam = None
tweakValue = None
# Determine which algorithm to use based on feedback
print "Running algorithm analyzer on QuestionID " + str(id)
if algorithm is None:
if info is None:
# If this is the first time running, choose one at random
print "Picking algorithm for the first time"
algorithms = {k: v for k, v in config.ALGORITHMS.iteritems() if v['Active'] is True}
algorithm = random.choice(algorithms.keys())
else:
# Need to check if algorithm picked is currently active in config file
# IF we need to switch algs based on info of ai model
if info['AIFeedback'] == '\x01': #If AIFeedback is true
print "AIFeedback is true"
algorithm = info['Algorithm']
else:
print info['AIFeedback']
print "AIFeedback is false, picking new algorithm"
algorithms = {k: v for k, v in config.ALGORITHMS.iteritems() if v['Active'] is True and k != info['Algorithm']}
algorithm = random.choice(algorithms.keys())
print "Algorithm chosen: " + algorithm
# Get parameters
latestAIModel = api.fetchLatestAIModelByAlgorithm(id, algorithm)
modelParams = []
modelParamsToSave = []
if latestAIModel is not None:
modelID = latestAIModel['ID']
modelParams = api.fetchAIModelParams(modelID)
print "Model ID " + str(modelID)
print modelParams
# else:
# Pick an optimizer if none was passed
if optimizer is None:
optimizers = {k: v for k, v in config.ALGORITHMS[algorithm]['Optimizers'].iteritems()}
if len(optimizers) > 0:
optimizer = random.choice(optimizers.keys())
print 'Optimizer chosen: ' + optimizer
# Pick a parameter from that optimizer
tweaks = {k: v for k, v in config.ALGORITHMS[algorithm]['Optimizers'][optimizer].iteritems()}
if len(tweaks) > 0:
tweak = random.choice(tweaks.keys())
if tweak is not None:
tweakParam = config.ALGORITHMS[algorithm]['Optimizers'][optimizer][tweak]['param']
tweakValue = config.ALGORITHMS[algorithm]['Optimizers'][optimizer][tweak]['value']
print "Tweak param " + str(tweakParam)
print "Tweak value " + str(tweakValue)
newParam = AIParam(None, tweakParam, tweakValue, optimizer)
# Add this param only if it isn't already added
existingParam = next((x for x in modelParams if x.Param == newParam.Param and x.param_use == newParam.param_use), None)
if existingParam is None and tweakParam is not None and tweakValue is not None:
modelParams.append(newParam)
# If we are using options, only pass options to classifier
if optimizer == 'Options':
for mParam in modelParams:
# I think this part is wrong, the truthiness of the downloaded from DB Optional Params I was unable to figure out
# PLEASE FIX
if mParam.param_use == 'Options' and (mParam.Value == True or mParam.Value == 1 or mParam.Value == "1"):
modelParamsToSave.append(mParam)
parameters = parameters + [mParam.Param]
# If we are using cvparams, only pass cvparams to classifier
elif optimizer == 'CVParams':
for mParam in modelParams:
if mParam.param_use == 'CVParams' and mParam.Value is not None:
modelParamsToSave.append(mParam)
parameters = parameters + ["-P", (mParam.Param + ' ' + mParam.Value)]
# If we are using ensemble, only pass ensemble params to classifier
elif optimizer == 'Ensemble':
for mParam in modelParams:
if mParam.param_use == 'Ensemble' and mParam.Value is not None:
modelParamsToSave.append(mParam)
parameters = parameters + [mParam.Param, mParam.Value]
# Run
instance = None
if optimizer == 'Options':
if algorithm == "SMO":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.functions.SMO', parameters, modelParamsToSave, optimizer)
elif algorithm == "RandomForest":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.trees.RandomForest', parameters, modelParamsToSave, optimizer)
elif algorithm == "NaiveBayes":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.bayes.NaiveBayes', parameters, modelParamsToSave, optimizer)
elif algorithm == "J48":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.trees.J48', parameters, modelParamsToSave, optimizer)
elif algorithm == "Perceptron":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.functions.MultilayerPerceptron', parameters, modelParamsToSave, optimizer)
elif optimizer == 'CVParams':
if algorithm == "SMO":
parameters = parameters + ["-W", "weka.classifiers.functions.SMO"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer)
elif algorithm == "RandomForest":
parameters = parameters + ["-W", "weka.classifiers.trees.RandomForest"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-W", "weka.classifiers.bayes.NaiveBayes"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer)
elif algorithm == "J48":
parameters = parameters + ["-W", "weka.classifiers.trees.J48"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer)
elif algorithm == "Perceptron":
parameters = parameters + ["-W", "weka.classifiers.functions.MultilayerPerceptron"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer)
elif optimizer == 'FeatureSelection':
if algorithm == "SMO":
parameters = parameters + ["-W", "weka.classifiers.functions.SMO", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer)
elif algorithm == "RandomForest":
parameters = parameters + ["-W", "weka.classifiers.trees.RandomForest", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-W", "weka.classifiers.bayes.NaiveBayes", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer)
elif algorithm == "J48":
parameters = parameters + ["-W", "weka.classifiers.trees.J48", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer)
elif algorithm == "Perceptron":
parameters = parameters + ["-W", "weka.classifiers.functions.MultilayerPerceptron", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer)
elif optimizer == 'Ensemble':
if algorithm == "SMO":
parameters = parameters + ["-B", "weka.classifiers.functions.SMO"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer)
elif algorithm == "RandomForest":
parameters = parameters + ["-B", "weka.classifiers.trees.RandomForest"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-B", "weka.classifiers.bayes.NaiveBayes"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer)
elif algorithm == "J48":
parameters = parameters + ["-B", "weka.classifiers.trees.J48"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer)
elif algorithm == "Perceptron":
parameters = parameters + ["-B", "weka.classifiers.functions.MultilayerPerceptron"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer)
return instance
def predict(id):
print "Running prediction on QuestionID " + str(id)
api = nemoApi()
config = nemoConfig()
# Fetch question info
info = api.fetchQuestionInfo(id)
# Declare vars to collect
algorithm = None
parameters = []
options = None
optimizer = None
# Get parameters
latestAIModel = api.fetchBestAIModelByQuestion(id)
modelParams = []
modelParamsToSave = []
if latestAIModel is not None:
modelID = latestAIModel['ID']
modelParams = api.fetchAIModelParams(modelID)
print "Model ID " + str(modelID)
print modelParams
algorithm = latestAIModel['Algorithm']
optimizer = latestAIModel['Optimizer']
# If we are using options, only pass options to classifier
if optimizer == 'Options':
for mParam in modelParams:
# I think this part is wrong, the truthiness of the downloaded from DB Optional Params I was unable to figure out
# PLEASE FIX
if mParam.param_use == 'Options' and (mParam.Value == True or mParam.Value == 1 or mParam.Value == "1"):
modelParamsToSave.append(mParam)
parameters = parameters + [mParam.Param]
# If we are using cvparams, only pass cvparams to classifier
elif optimizer == 'CVParams':
for mParam in modelParams:
if mParam.param_use == 'CVParams' and mParam.Value is not None:
modelParamsToSave.append(mParam)
parameters = parameters + ["-P", (mParam.Param + ' ' + mParam.Value)]
# If we are using ensemble, only pass ensemble params to classifier
elif optimizer == 'Ensemble':
for mParam in modelParams:
if mParam.param_use == 'Ensemble' and mParam.Value is not None:
modelParamsToSave.append(mParam)
parameters = parameters + [mParam.Param, mParam.Value]
# Run
instance = None
if optimizer == 'Options':
if algorithm == "SMO":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.functions.SMO', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "RandomForest":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.trees.RandomForest', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "NaiveBayes":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.bayes.NaiveBayes', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "J48":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.trees.J48', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "Perceptron":
instance = WekaWrapper(id, algorithm, 'weka.classifiers.functions.MultilayerPerceptron', parameters, modelParamsToSave, optimizer, predict=1)
elif optimizer == 'CVParams':
if algorithm == "SMO":
parameters = parameters + ["-W", "weka.classifiers.functions.SMO"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "RandomForest":
parameters = parameters + ["-W", "weka.classifiers.trees.RandomForest"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-W", "weka.classifiers.bayes.NaiveBayes"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "J48":
parameters = parameters + ["-W", "weka.classifiers.trees.J48"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "Perceptron":
parameters = parameters + ["-W", "weka.classifiers.functions.MultilayerPerceptron"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.CVParameterSelection', parameters, modelParamsToSave, optimizer, predict=1)
elif optimizer == 'FeatureSelection':
if algorithm == "SMO":
parameters = parameters + ["-W", "weka.classifiers.functions.SMO", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "RandomForest":
parameters = parameters + ["-W", "weka.classifiers.trees.RandomForest", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-W", "weka.classifiers.bayes.NaiveBayes", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "J48":
parameters = parameters + ["-W", "weka.classifiers.trees.J48", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "Perceptron":
parameters = parameters + ["-W", "weka.classifiers.functions.MultilayerPerceptron", "-E", "weka.attributeSelection.CfsSubsetEval -M", "-S", "weka.attributeSelection.BestFirst -D 1 -N 5"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.AttributeSelectedClassifier', parameters, modelParamsToSave, optimizer, predict=1)
elif optimizer == 'Ensemble':
if algorithm == "SMO":
parameters = parameters + ["-B", "weka.classifiers.functions.SMO"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "RandomForest":
parameters = parameters + ["-B", "weka.classifiers.trees.RandomForest"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "NaiveBayes":
parameters = parameters + ["-B", "weka.classifiers.bayes.NaiveBayes"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "J48":
parameters = parameters + ["-B", "weka.classifiers.trees.J48"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer, predict=1)
elif algorithm == "Perceptron":
parameters = parameters + ["-B", "weka.classifiers.functions.MultilayerPerceptron"]
instance = WekaWrapper(id, algorithm, 'weka.classifiers.meta.Stacking', parameters, modelParamsToSave, optimizer, predict=1)
return instance
else:
return None