def ridgereport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# Todo: get Logistic feature selection working
("randlogit", RandomizedLogisticRegression()),
("logit", LogisticRegression())
])
algorithm = "Logistic_Regression"
baseparam = "{'logit__C': (np.logspace(-2, 2, 10))"
elif self.modeltype == 'regress':
algorithm = Ridge()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("regress", LinearRegression())
baseparam = {'regress__alpha': (np.logspace(-2, 2, 10)),
'regress__normalize': (True, False)}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
#Todo: make plotting symmetric between regress/class
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="RidgeRegression")
def randomforestreport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
# self.pipeline = Pipeline([("imputer", Imputer(
# axis=0)),
# ("feature_selection", SelectFromModel(
# LinearSVC(), threshold="median")),
# ("randforest", RandomForestClassifier())])
algorithm = RandomForestClassifier()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
baseparam = {'classify__n_estimators': [10,50,100,250,500]}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
elif self.modeltype == "regress":
# todo: Add auto feature selection into auto pipeline creation
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# #("feature_selection", RFE(
# # RandomForestRegressor(), 6)),
# ("regress", RandomForestRegressor())])
algorithm = RandomForestRegressor()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
baseparam = {'regress__n_estimators': [10,50,100,250,500]}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("regress", RandomForestRegressor())
# ])
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="RandomForest")
def logitreport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
algorithm = LogisticRegression()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# ("randlogit", RandomizedLogisticRegression()),
# ("logit", LogisticRegression())
# ])
baseparam = {'classify__C': (np.logspace(-2, 2, 10))}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
elif self.modeltype == 'regress':
algorithm = LinearRegression()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("regress", LinearRegression())
baseparam = {'regress__normalize': (False, True)}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
#Todo: make plotting symmetric between regress/class
#Todo: make plotit, saveit False everywhere by default (such that they're optional parameters)
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="LinearRegression")
def treereport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
self.pipeline = Pipeline([("imputer", Imputer(
axis=0)),
("feature_selection", SelectFromModel(
LinearSVC(), threshold="median")),
("trees", DecisionTreeClassifier())])
title = "Decision_Tree_Classifier"
#algorithm = DecisionTreeClassifier()
#self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
# Set the parameters by cross-validation
self.parameters = {'trees__criterion': ["gini", "entropy"],
'trees__class_weight': ["balanced"],
'imputer__strategy': ('mean', 'median', 'most_frequent')}
elif self.modeltype == "regress":
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("regress", DecisionTreeRegressor())
# ])
algorithm = DecisionTreeRegressor()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
baseparam = {'regress__splitter': ('best','random')}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="DecisionTree")
def gradboostreport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
pass
#todo: set up classifier gridsearch
elif self.modeltype == "regress":
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# #("feature_selection", RFE(
# # RandomForestRegressor(), 6)),
# ("regress", RandomForestRegressor())])
algorithm = GradientBoostingRegressor()
#
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
#
baseparam = {'regress__loss': ['ls','lad'],
#'regress__learning_rate': [0.05, 0.1],
'regress__n_estimators': [75,100,250,350,500]}
#
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("regress", RandomForestRegressor())
# ])
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="GradientBoosting")
def svmreport(self, folds, cores, plotit, saveit):
self.folds = folds
self.cores = cores
if self.modeltype == 'classify':
self.pipeline = Pipeline([("imputer", Imputer(
axis=0)),
("feature_selection", SelectFromModel(
LinearSVC(), threshold="median")),
("lsvc", LinearSVC())])
# Set the parameters by cross-validation
self.parameters = {'imputer__strategy': ('mean', 'median', 'most_frequent'),
'lsvc__C': (np.logspace(-2, 2, 10))}
title = "Linear_SVC"
elif self.modeltype == "regress":
# self.pipeline = Pipeline([("imputer", Imputer(axis=0)),
# # Todo: get Logistic feature selection working (test on IU)
# #("randlogit", RandomizedLogisticRegression()),
# ("lsvr", LinearSVR())])
algorithm = LinearSVR()
self.pipeline = modelutilities.buildclfpipeline(algorithm, self.impute, self.modeltype)
baseparam = {'regress__C': (np.logspace(-2, 2, 10))}
self.parameters = modelutilities.buildgridparameters(baseparam, self.impute, self.modeltype)
# todo: fix plt.show() stopping program execution
# Fit/Predict, run report, and return fit
return TuneModel.clfreport(self, printsample=True, plotit=plotit, saveit=saveit, title="SVM")
