def Execute(context):
timeStamp = TimeStamp()
connectionSQL = ConnectionSQL(context)
criterion = 'gini'
max_features = None
max_depth = 14
min_samples_split=6
min_samples_leaf=6
n_jobs=-1
# model parameter ranges - used for searching for optimial parameter settings
criterionList = ['gini', 'entropy']
max_featuresList = [7, 8, 9, 10, 11, 12, 13, 14, None]
max_depthList = [6, 8, 10, 11, 12, 13, 14, None]
min_samples_splitList = [3, 5, 6, 7, 8, 9]
min_samples_leafList = [3, 4, 5, 6, 7]
parameterRangeDict = {'criterion':criterionList, 'max_features':max_featuresList, 'max_depth':max_depthList, 'min_samples_split':min_samples_splitList, 'min_samples_leaf':min_samples_leafList }
timeStamp1 = TimeStamp('Loading X, Y')
# 0=DataID, 1=P_A, 2=P_B, 3=P_C, 4=P_D, 5=P_E, 6=P_F, 7=P_G, 8=State, ... 27=Cost
# Load Training Data and Cross Validation Data
train_X, train_Y, train_DataID = connectionSQL.GetFeaturesAndResultsFromCache(featuresColumns=__featuresColumns, dataType="Train", modelName=__modelName, preProcessDataFrame=__PreProcessDataFrame, viewName=__viewName)
train_DataID = None
# Load Cross Validation Data
cross_X, cross_Y, cross_DataID = connectionSQL.GetFeaturesAndResultsFromCache(featuresColumns=__featuresColumns, dataType="Cross", modelName=__modelName, preProcessDataFrame=__PreProcessDataFrame, viewName=__viewName)
cross_DataID = None
print ' Elaspe=' + timeStamp1.Elaspse
print ''
if (__optimiseParameters):
timeStamp1 = TimeStamp('Optimise parameters')
# package paramaters
dataDict = {'train_X':train_X, 'train_Y':train_Y, 'cross_X':cross_X, 'cross_Y':cross_Y}
# Optimal Paramater selection
#parameterDictMax = __OptimalParamaterSelection(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.2 # 20%
parameterDictMax, accuracyMax = optimiseModelParameters.ExecuteMoneCarlo(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
criterion = parameterDictMax['criterion']
max_features = parameterDictMax['max_features']
max_depth = parameterDictMax['max_depth']
min_samples_split = parameterDictMax['min_samples_split']
min_samples_leaf = parameterDictMax['min_samples_leaf']
print 'Elaspe=' + timeStamp1.Elaspse
print ''
timeStamp1 = TimeStamp('Fit model')
print __modelName + ' criterion=' + criterion + ' max_features=' + str(max_features) + ' max_depth=' + str(max_depth) + ' min_samples_split=' + str(min_samples_split) + ' min_samples_leaf=' + str(min_samples_leaf)
clf = DecisionTreeClassifierMultiClass(splitter='best', random_state=1,
criterion=criterion, max_features=max_features, max_depth=max_depth, min_samples_split=min_samples_split, min_samples_leaf=min_samples_leaf)
model = clf.fit(train_X, train_Y) # All features must be float.
print 'Elaspe=' + timeStamp1.Elaspse
print ''
# Training error reports
SharedLibrary.TrainingError(dataType='Training', modelName=__modelName, clf=clf, model=model, dfX=train_X, dfY=train_Y)
SharedLibrary.TrainingError(dataType='CrossVal', modelName=__modelName, clf=clf, model=model, dfX=cross_X, dfY=cross_Y)
# Cross validation report
SharedLibrary.CrossValidation(modelName=__modelName, clf=clf, dfX=train_X, dfY=train_Y, n_fold=5)
# UPDATE DRV_Predict with all data predictions.
timeStamp1 = TimeStamp('Update_DRV_Predict')
all_X, all_Y, all_DataID = connectionSQL.GetFeaturesAndResultsFromCache(featuresColumns=__featuresColumns, dataType="ALL", modelName=__modelName, preProcessDataFrame=__PreProcessDataFrame, viewName=__viewName)
all_Y = None
connectionSQL.Update_DRV_Predict(modelName=__modelName, model=model, all_X=all_X, all_DataID=all_DataID)
print 'Elaspe=' + timeStamp1.Elaspse
print ''
# Feature analysis summary
featureImportances = enumerate(clf.feature_importances_)
featureImportanceHistogram = np.array([(importance,train_X.columns[i]) for (i,importance) in featureImportances if importance > 0.01])
print __modelName + ' Elaspe=' + timeStamp.Elaspse
return __modelName, featureImportanceHistogram
def Execute(context):
connectionSQL = ConnectionSQL(context)
penalty = None
alpha = 0.0001
fit_intercept = True
n_iter = 20
shuffle = False
eta0 = 1
# model parameter ranges - used for searching for optimial parameter settings
penaltyList = ['l1', 'l2', 'elasticnet', None]
alphaList = [0.1, 0.001, 0.0001, 0.00001]
fit_interceptList = [True, False]
n_iterList = [5, 10, 20, 30, 40, 100]
shuffleList = [False]
eta0List = [1]
parameterRangeDict = {'penalty':penaltyList, 'alpha':alphaList, 'fit_intercept':fit_interceptList, 'n_iter':n_iterList, 'shuffle':shuffleList, 'eta0':eta0List }
# 0=DataID, 1=Actual, 2=Base, ...
# Load Training Data and Cross Validation Data
train_X, train_Y, train_DataID = __GetFeaturesAndResults(context=context, connectionSQL=connectionSQL, dataType="Train", preProcessDataFrame=__PreProcessDataFrame)
train_DataID = None
# Load Cross Validation Data
cross_X, cross_Y, cross_DataID = __GetFeaturesAndResults(context=context, connectionSQL=connectionSQL, dataType="Cross", preProcessDataFrame=__PreProcessDataFrame)
cross_DataID = None
if (__optimiseParameters):
parameterDictMax = {}
accuracyMax = 0.0
# package paramaters
dataDict = {'train_X':train_X, 'train_Y':train_Y, 'cross_X':cross_X, 'cross_Y':cross_Y}
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.20 # 20%
parameterDictMax, accuracyMax = optimiseModelParameters.ExecuteMoneCarlo(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
penalty = parameterDictMax['penalty']
alpha = parameterDictMax['alpha']
fit_intercept = parameterDictMax['fit_intercept']
n_iter = parameterDictMax['n_iter']
shuffle = parameterDictMax['shuffle']
eta0 = parameterDictMax['eta0']
print __modelName + 'Classifier'
clf = Perceptron(penalty=penalty, alpha=alpha, fit_intercept=fit_intercept, n_iter=n_iter, shuffle=shuffle, random_state=1, eta0=eta0, warm_start=False)
model = clf.fit(train_X, train_Y) # All features must be float.
print ''
# Training error reports
SharedLibrary.TrainingError(dataType='Training', modelName=__modelName, clf=clf, model=model, dfX=train_X, dfY=train_Y)
SharedLibrary.TrainingError(dataType='CrossVal', modelName=__modelName, clf=clf, model=model, dfX=cross_X, dfY=cross_Y)
# Cross validation report
SharedLibrary.CrossValidation(modelName=__modelName, clf=clf, dfX=train_X, dfY=train_Y, n_fold=5)
weights = clf.coef_[0] # narray
weights = weights/sum(weights)
modelNames = train_X.keys();
weightsDict = {}
for index in range(len(modelNames)):
weightsDict[modelNames[index]] = weights[index]
weightsString = ''
prefix = ''
for key, value in weightsDict.iteritems():
weightsString += prefix + key + '=' + str(value)
prefix = ' '
print 'Weights ' + weightsString
print ''
# UPDATE DRV_Predict with all data predictions.
all_X, all_Y, all_DataID = __GetFeaturesAndResults(context=context, connectionSQL=connectionSQL, dataType="ALL", preProcessDataFrame=__PreProcessDataFrame)
all_Y = None
connectionSQL.Update_DRV_Predict(modelName=__modelName, model=model, all_X=all_X, all_DataID=all_DataID)
return __modelName
def Execute(context):
connectionSQL = ConnectionSQL(context)
kernel = 'linear'
degree = 3
# model parameter ranges - used for searching for optimial parameter settings
kernelList = ['linear','poly','rbf','sigmoid']
parameterRangeDict = {'kernel':kernelList }
# 0=DataID, 1=Actual, 2=Pclass, ...
# Load Training Data and Cross Validation Data
train_X, train_Y, train_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="Train", preProcessDataFrame=__PreProcessDataFrame)
train_DataID = None
# Load Cross Validation Data
cross_X, cross_Y, cross_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="Cross", preProcessDataFrame=__PreProcessDataFrame)
cross_DataID = None
if (__optimiseParameters):
parameterDictMax = {}
accuracyMax = 0.0
# package paramaters
dataDict = {'train_X':train_X, 'train_Y':train_Y, 'cross_X':cross_X, 'cross_Y':cross_Y}
# Optimal Paramater selection
# Kernel=Linear
kernelList = ['linear']
degreeList = [3]
parameterRangeDict = {'kernel':kernelList, 'degree':degreeList }
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.2 # 20%
parameterDict, accuracy = optimiseModelParameters.ExecuteExhaustive(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
if (accuracy > accuracyMax):
parameterDictMax = parameterDict
accuracyMax = accuracy
# Kernel=poly
kernelList = ['poly']
degreeList = [2, 3, 4]
parameterRangeDict = {'kernel':kernelList, 'degree':degreeList }
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.2 # 20%
parameterDict, accuracy = optimiseModelParameters.ExecuteExhaustive(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
if (accuracy > accuracyMax):
parameterDictMax = parameterDict
accuracyMax = accuracy
# Kernel=rbf
kernelList = ['rbf']
degreeList = [3]
parameterRangeDict = {'kernel':kernelList, 'degree':degreeList }
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.2 # 20%
parameterDict, accuracy = optimiseModelParameters.ExecuteExhaustive(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
if (accuracy > accuracyMax):
parameterDictMax = parameterDict
accuracyMax = accuracy
# Kernel=sigmoid
kernelList = ['sigmoid']
degreeList = [3]
parameterRangeDict = {'kernel':kernelList, 'degree':degreeList }
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.2 # 20%
parameterDict, accuracy = optimiseModelParameters.ExecuteExhaustive(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
if (accuracy > accuracyMax):
parameterDictMax = parameterDict
accuracyMax = accuracy
kernel = parameterDictMax['kernel']
degree = parameterDictMax['degree']
print __modelName + 'Classifier'
clf = svm.SVC(probability=True, verbose=False, random_state=1, kernel=kernel, degree=degree)
model = clf.fit(train_X, train_Y) # All features must be float.
print ''
# Training error reports
SharedLibrary.TrainingError(dataType='Training', modelName=__modelName, clf=clf, model=model, dfX=train_X, dfY=train_Y)
SharedLibrary.TrainingError(dataType='CrossVal', modelName=__modelName, clf=clf, model=model, dfX=cross_X, dfY=cross_Y)
# Cross validation report
SharedLibrary.CrossValidation(modelName=__modelName, clf=clf, dfX=train_X, dfY=train_Y, n_fold=5)
# UPDATE DRV_Predict with all data predictions.
all_X, all_Y, all_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="ALL", preProcessDataFrame=__PreProcessDataFrame)
all_Y = None
connectionSQL.Update_DRV_Predict(modelName=__modelName, model=model, all_X=all_X, all_DataID=all_DataID)
return __modelName
def Execute(context):
connectionSQL = ConnectionSQL(context)
# Default model parameter values
n_estimators = 50
max_features = 8 # None
max_depth = 12
min_samples_split=5
min_samples_leaf=4
bootstrap=False
n_jobs=-1 # Avaiable CPUs
random_state=1 # Use None or 0 to randomise results
# model parameter ranges - used for searching for optimial parameter settings
n_estimatorsList = [20, 30, 40, 50, 70, 80, 100, 110]
max_featuresList = [7, 8, 9, 10, 11, 12, 13, 14, None]
#max_featuresList = ['auto']
max_depthList = [6, 8, 10, 11, 12, 13, 14]
min_samples_splitList = [3, 5, 6, 7, 8, 9]
min_samples_leafList = [3, 4, 5, 6, 7]
bootstrapList = [False]
parameterRangeDict = {'n_estimators':n_estimatorsList, 'max_features':max_featuresList, 'max_depth':max_depthList, 'min_samples_split':min_samples_splitList, 'min_samples_leaf':min_samples_leafList, 'bootstrap':bootstrapList }
# 0=DataID, 1=Actual, 2=Pclass, ...
# Load Training Data and Cross Validation Data
train_X, train_Y, train_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="Train", preProcessDataFrame=__PreProcessDataFrame)
#train_DataID = None
# Load Cross Validation Data
cross_X, cross_Y, cross_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="Cross", preProcessDataFrame=__PreProcessDataFrame)
#cross_DataID = None
if (__optimiseParameters):
# package paramaters
dataDict = {'train_X':train_X, 'train_Y':train_Y, 'cross_X':cross_X, 'cross_Y':cross_Y}
# Optimal Paramater selection
#parameterDictMax = __OptimalParamaterSelection(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
optimiseModelParameters = OptimiseModelParameters(modelName=__modelName)
optimiseModelParameters.Percent = 0.05 # 5%
parameterDictMax, accuracyMax = optimiseModelParameters.ExecuteMoneCarlo(dataDict=dataDict, accuracyFunct=__Accuracy, parameterRangeDict=parameterRangeDict)
bootstrap = parameterDictMax['bootstrap']
min_samples_leaf = parameterDictMax['min_samples_leaf']
n_estimators = parameterDictMax['n_estimators']
max_features = parameterDictMax['max_features']
min_samples_split = parameterDictMax['min_samples_split']
max_depth = parameterDictMax['max_depth']
print __modelName + ' bootstrap=' + str(bootstrap) + ' min_samples_leaf=' + str(min_samples_leaf) + ' n_estimators=' + str(n_estimators) + " max_features=" + str(max_features) + " min_samples_split=" + str(min_samples_split) + " max_depth=" + str(max_depth)
oob_score = bootstrap # Can only have oob_score if bootstrap = True
clf = RandomForestClassifier(n_jobs=-1, oob_score=oob_score, random_state=1,
bootstrap=bootstrap, min_samples_leaf=min_samples_leaf, n_estimators=n_estimators, max_features=max_features, min_samples_split=min_samples_split, max_depth=max_depth)
model = clf.fit(train_X, train_Y) # All features must be float.
print ''
# Persist trained classifier to disk
# clfUNC = __modelName + '.clf'
#joblib.dump(clfUNC)
#clf = joblib.load(clfUNC)
# Training error reports
SharedLibrary.TrainingError(dataType='Training', modelName=__modelName, clf=clf, model=model, dfX=train_X, dfY=train_Y)
SharedLibrary.TrainingError(dataType='CrossVal', modelName=__modelName, clf=clf, model=model, dfX=cross_X, dfY=cross_Y)
# Cross validation report
SharedLibrary.CrossValidation(modelName=__modelName, clf=clf, dfX=train_X, dfY=train_Y, n_fold=5)
# UPDATE DRV_Predict with all data predictions.
all_X, all_Y, all_DataID = connectionSQL.GetFeaturesAndResults(featuresColumns=__featuresColumns, dataType="ALL", preProcessDataFrame=__PreProcessDataFrame)
all_Y = None
connectionSQL.Update_DRV_Predict(modelName=__modelName, model=model, all_X=all_X, all_DataID=all_DataID)
# Feature analysis summary
featureImportances = enumerate(clf.feature_importances_)
featureImportanceHistogram = np.array([(importance,train_X.columns[i]) for (i,importance) in featureImportances if importance > 0.005])
return __modelName, featureImportanceHistogram
