def __init__(self, data):
# User defined arguments
fixed_random_seed = 0xbeef
train_data_sampler = CompleteData()
missing_value_handler = CompleteCaseAnalysis()
numeric_attribute_scaler = NoScaler()
learners = [
NonTunedLogisticRegression(),
NonTunedDecisionTree()
]
pre_processors = [NoPreProcessing()]
post_processors = [NoPostProcessing()]
# Fixed arguments for dataset
test_set_ratio = 0.2
validation_set_ratio = 0.1
label_name = 'credit'
positive_label = 1
numeric_attribute_names = [
'month', 'credit_amount', 'residence_since', 'age',
'number_of_credits', 'people_liable_for'
]
categorical_attribute_names = [
'credit_history', 'savings', 'employment'
]
attributes_to_drop_names = [
'personal_status', 'status', 'purpose',
'investment_as_income_percentage', 'other_debtors',
'property', 'installment_plans', 'housing', 'skill_level',
'telephone', 'foreign_worker'
]
protected_attribute_names = ['sex']
privileged_classes = [[1.0]]
privileged_groups = [{'sex': 1.0}]
unprivileged_groups = [{'sex': 0.0}]
dataset_metadata = {
'label_maps': [{
1.0: 1,
0.0: 0
}],
'protected_attribute_maps': [{
1.0: 'male',
0.0: 'female'
}]
}
dataset_name = 'test_dataset'
# Constructor call
super().__init__(
fixed_random_seed, test_set_ratio, validation_set_ratio,
label_name, positive_label, numeric_attribute_names,
categorical_attribute_names, attributes_to_drop_names,
train_data_sampler, missing_value_handler,
numeric_attribute_scaler, learners, pre_processors,
post_processors, protected_attribute_names,
privileged_classes, privileged_groups, unprivileged_groups,
dataset_metadata, dataset_name)
def test_GiveMeSomeCreditExperiment(self):
self.experiment = GiveMeSomeCreditExperiment(
fixed_random_seed=0xabcd,
train_data_sampler=CompleteData(),
missing_value_handler=CompleteCaseAnalysis(),
numeric_attribute_scaler=NamedStandardScaler(),
learners=[NonTunedLogisticRegression(),
NonTunedDecisionTree()],
pre_processors=[NoPreProcessing()],
post_processors=[NoPostProcessing()])
self.experiment.run()
from fp.dataset_experiments import GermanCreditDatasetSexExperiment
from fp.scalers import NamedStandardScaler, NamedMinMaxScaler
from fp.learners import NonTunedLogisticRegression, LogisticRegression, DecisionTree, NonTunedLogisticRegression, NonTunedDecisionTree
from fp.post_processors import NoPostProcessing, RejectOptionPostProcessing, EqualOddsPostProcessing, CalibratedEqualOddsPostProcessing
from fp.pre_processors import NoPreProcessing, DIRemover, Reweighing
import numpy as np
import matplotlib.pyplot as plt
#creating list of parameters that we will alter to observe variations
seeds = [
0xbeef, 0xcafe, 0xdead, 0xdeadcafe, 0xdeadbeef, 0xbeefcafe, 0xcafebeef, 50,
583, 5278, 100000, 0xefac, 0xfeeb, 0xdaed, 0xefacdaed, 0xfeebdead
]
learners = [
NonTunedLogisticRegression(),
LogisticRegression(),
NonTunedDecisionTree(),
DecisionTree()
]
processors = [(NoPreProcessing(), NoPostProcessing()),
(DIRemover(1.0), NoPostProcessing()),
(DIRemover(0.5), NoPostProcessing()),
(Reweighing(), NoPostProcessing()),
(NoPreProcessing(), RejectOptionPostProcessing()),
(NoPreProcessing(), CalibratedEqualOddsPostProcessing())]
def calculate_metrics(seed, learner, pre_processor, post_processor):
'''
Experiment function to run the experiments
def test_NonTunedLogisticRegression(self):
self.learner = NonTunedLogisticRegression()
result = self.learner.fit_model(self.annotated_data, self.seed)
self.assertEqual(self.learner.name(), 'LogisticRegression-notuning')
self.assertEqual(type(result), type(SGDClassifier()))