class SvcConfiguration(ClassifierConfiguration):
def __init__(self, num_folds, sample_weight, families_supervision,
test_conf):
ClassifierConfiguration.__init__(self,
num_folds,
sample_weight,
families_supervision,
test_conf=test_conf)
self.model_class = Svc
self.c = LearningParameter(list(10.**np.arange(-2, 2)))
def getModelClassName(self):
return 'Svc'
def setC(self, c_values):
self.c = LearningParameter(c_values)
def getParamGrid(self):
param_grid = {'model__C': self.c.values}
return param_grid
def setBestValues(self, grid_search):
self.c.setBestValue(grid_search.best_params_['model__C'])
def getBestValues(self):
best_values = {'model__C': self.c.best_value}
return best_values
@staticmethod
def fromJson(obj, exp):
test_conf = TestConfiguration.fromJson(obj['test_conf'], exp)
conf = SvcConfiguration(obj['num_folds'], obj['sample_weight'],
obj['families_supervision'], test_conf)
conf.c = LearningParameter.fromJson(obj['c'])
return conf
def toJson(self):
conf = ClassifierConfiguration.toJson(self)
conf['__type__'] = 'SvcConfiguration'
conf['c'] = self.c.toJson()
return conf
def probabilistModel(self):
return False
def semiSupervisedModel(self):
return False
def featureImportance(self):
return None
@staticmethod
def generateParser(parser):
classifier_group = ClassifierConfiguration.generateParser(parser)
@staticmethod
def generateParamsFromArgs(args, experiment):
params = ClassifierConfiguration.generateParamsFromArgs(
args, experiment)
return params
def __init__(self, num_folds, sample_weight, families_supervision,
test_conf):
ClassifierConfiguration.__init__(self,
num_folds,
sample_weight,
families_supervision,
test_conf=test_conf)
self.model_class = Svc
self.c = LearningParameter(list(10.**np.arange(-2, 2)))
def fromJson(obj, exp):
test_conf = TestConfiguration.fromJson(obj['test_conf'], exp)
conf = LogisticRegressionConfiguration(obj['num_folds'], obj['sample_weight'],
obj['families_supervision'], obj['optim_algo'],
test_conf)
conf.c = LearningParameter.fromJson(obj['c'])
conf.penalty = LearningParameter.fromJson(obj['penalty'])
conf.optim_algo = obj['optim_algo']
return conf
def __init__(self, num_folds, sample_weight, families_supervision,
test_conf):
ClassifierConfiguration.__init__(self,
num_folds,
sample_weight,
families_supervision,
test_conf=test_conf)
self.model_class = LogisticRegression
#self.optim_algo = 'sag'
#self.c = LearningParameter(list(10. ** np.arange(-2, 2)))
#self.penalty = LearningParameter(['l2'])
self.optim_algo = 'liblinear'
self.c = LearningParameter(list(10.**np.arange(-2, 2)))
self.penalty = LearningParameter(['l1', 'l2'])
def fromJson(obj, exp):
conf = SvcConfiguration(obj['num_folds'], obj['sample_weight'],
obj['families_supervision'])
ClassifierConfiguration.setTestConfiguration(conf, obj, exp)
conf.c = LearningParameter.fromJson(obj['c'])
return conf
def setC(self, c_values):
self.c = LearningParameter(c_values)
def setPenalty(self, penalty_values):
self.penalty = LearningParameter(penalty_values)
class LogisticRegressionConfiguration(ClassifierConfiguration):
def __init__(self, num_folds, sample_weight, families_supervision,
optim_algo, test_conf):
ClassifierConfiguration.__init__(self,
num_folds,
sample_weight,
families_supervision,
test_conf=test_conf)
self.model_class = LogisticRegression
if optim_algo is not None:
self.optim_algo = optim_algo
else:
self.optim_algo = 'liblinear'
self.c = LearningParameter(list(10.**np.arange(-2, 2)))
if self.optim_algo == 'sag':
self.penalty = LearningParameter(['l2'])
elif self.optim_algo == 'liblinear':
self.penalty = LearningParameter(['l1', 'l2'])
def getModelClassName(self):
return 'LogisticRegression'
def generateSuffix(self):
suffix = ClassifierConfiguration.generateSuffix(self)
suffix += '__' + self.optim_algo
return suffix
def setC(self, c_values):
self.c = LearningParameter(c_values)
def setPenalty(self, penalty_values):
self.penalty = LearningParameter(penalty_values)
def setOptimAlgo(self, optim_algo):
self.optim_algo = optim_algo
def getParamGrid(self):
param_grid = {
'model__C': self.c.values,
'model__penalty': self.penalty.values
}
return param_grid
def setBestValues(self, grid_search):
self.c.setBestValue(grid_search.best_params_['model__C'])
self.penalty.setBestValue(grid_search.best_params_['model__penalty'])
def getBestValues(self):
best_values = {
'model__C': self.c.best_value,
'model__penalty': self.penalty.best_value
}
return best_values
@staticmethod
def fromJson(obj, exp):
test_conf = TestConfiguration.fromJson(obj['test_conf'], exp)
conf = LogisticRegressionConfiguration(obj['num_folds'],
obj['sample_weight'],
obj['families_supervision'],
obj['optim_algo'], test_conf)
ClassifierConfiguration.setTestConfiguration(conf, obj, exp)
conf.c = LearningParameter.fromJson(obj['c'])
conf.penalty = LearningParameter.fromJson(obj['penalty'])
conf.optim_algo = obj['optim_algo']
return conf
def toJson(self):
conf = ClassifierConfiguration.toJson(self)
conf['__type__'] = 'LogisticRegressionConfiguration'
conf['optim_algo'] = self.optim_algo
conf['c'] = self.c.toJson()
conf['penalty'] = self.penalty.toJson()
return conf
def probabilistModel(self):
return True
def semiSupervisedModel(self):
return False
def featureCoefficients(self):
return not (self.families_supervision)
@staticmethod
def generateParser(parser):
ClassifierConfiguration.generateParser(parser)
parser.add_argument('--optim-algo',
choices=['sag', 'liblinear'],
default='liblinear',
help='sag is recommended for large datasets.')
@staticmethod
def generateParamsFromArgs(args, experiment):
params = ClassifierConfiguration.generateParamsFromArgs(
args, experiment)
params['optim_algo'] = args.optim_algo
return params
def fromJson(obj, exp):
test_conf = TestConfiguration.fromJson(obj['test_conf'], exp)
conf = SvcConfiguration(obj['num_folds'], obj['sample_weight'],
obj['families_supervision'], test_conf)
conf.c = LearningParameter.fromJson(obj['c'])
return conf
class LogisticRegressionConfiguration(ClassifierConfiguration):
def __init__(self, num_folds, sample_weight, families_supervision,
test_conf):
ClassifierConfiguration.__init__(self,
num_folds,
sample_weight,
families_supervision,
test_conf=test_conf)
self.model_class = LogisticRegression
#self.optim_algo = 'sag'
#self.c = LearningParameter(list(10. ** np.arange(-2, 2)))
#self.penalty = LearningParameter(['l2'])
self.optim_algo = 'liblinear'
self.c = LearningParameter(list(10.**np.arange(-2, 2)))
self.penalty = LearningParameter(['l1', 'l2'])
def getModelClassName(self):
return 'LogisticRegression'
def setC(self, c_values):
self.c = LearningParameter(c_values)
def setPenalty(self, penalty_values):
self.penalty = LearningParameter(penalty_values)
def setOptimAlgo(self, optim_algo):
self.optim_algo = optim_algo
def getParamGrid(self):
param_grid = {
'model__C': self.c.values,
'model__penalty': self.penalty.values
}
return param_grid
def setBestValues(self, grid_search):
self.c.setBestValue(grid_search.best_params_['model__C'])
self.penalty.setBestValue(grid_search.best_params_['model__penalty'])
def getBestValues(self):
best_values = {
'model__C': self.c.best_value,
'model__penalty': self.penalty.best_value
}
return best_values
@staticmethod
def fromJson(obj, exp):
test_conf = TestConfiguration.fromJson(obj['test_conf'], exp)
conf = LogisticRegressionConfiguration(obj['num_folds'],
obj['sample_weight'],
obj['families_supervision'],
test_conf)
ClassifierConfiguration.setTestConfiguration(conf, obj, exp)
conf.c = LearningParameter.fromJson(obj['c'])
conf.penalty = LearningParameter.fromJson(obj['penalty'])
conf.optim_algo = obj['optim_algo']
return conf
def toJson(self):
conf = ClassifierConfiguration.toJson(self)
conf['__type__'] = 'LogisticRegressionConfiguration'
conf['optim_algo'] = self.optim_algo
conf['c'] = self.c.toJson()
conf['penalty'] = self.penalty.toJson()
return conf
def probabilistModel(self):
return True
def semiSupervisedModel(self):
return False
def featureCoefficients(self):
return not (self.families_supervision)
@staticmethod
def generateParser(parser):
classifier_group = ClassifierConfiguration.generateParser(parser)
@staticmethod
def generateParamsFromArgs(args, experiment):
params = ClassifierConfiguration.generateParamsFromArgs(
args, experiment)
return params