def test_iris_one_student(self):
data = self.load_data()
oracle = Oracle(teacher=RuleModel(),
student_modelers=[RandomForestModeler()])
oracle.build(data['training_data'])
pred = oracle.predict(data['test_data'])['predictions']
assert len(pred) == len(data['test_data']['y'])
assert any(pred != data['test_data']['y'])
# print(metrics.f1_score(data['test_data']['y'], pred, average='macro'))
with tempfile.TemporaryDirectory() as path:
os.environ['H1ST_MODEL_REPO_PATH'] = path
version = oracle.persist()
oracle_2 = Oracle(teacher=RuleModel(),
student_modelers=[RandomForestModeler()])
oracle_2.load_params(version)
assert 'sklearn' in str(type(oracle_2.students[0].base_model))
pred_2 = oracle_2.predict(data['test_data'])['predictions']
pred_df = pd.concat((pred, pred_2), axis=1)
pred_df.columns = ['predictions', 'pred_2']
assert len(
pred_df[pred_df['predictions'] != pred_df['pred_2']]) == 0
def test_rule_based_ensemble_one_equipment(self):
data = self.load_data()
data['training_data']['X'].drop('machineID', axis=1, inplace=True)
oracle_modeler = TimeseriesOracleModeler(
teacher=RuleModel(),
student_modelers=[RandomForestModeler(),
AdaBoostModeler()],
ensembler_modeler=RuleBasedModeler(
model_class=RuleBasedClassificationModel))
oracle = oracle_modeler.build_model(
{'unlabeled_data': data['training_data']['X']}, ts_col='date')
pred = oracle.predict(data['training_data'])['predictions']
assert len(pred) == 2
with tempfile.TemporaryDirectory() as path:
os.environ['H1ST_MODEL_REPO_PATH'] = path
version = oracle.persist()
oracle_2 = TimeSeriesOracle(
teacher=RuleModel(),
students=[RandomForestModel(),
AdaBoostModel()],
ensembler=RuleBasedClassificationModel())
oracle_2.load_params(version)
assert 'sklearn' in str(type(oracle_2.students[0].base_model))
pred_2 = oracle_2.predict(data['training_data'])['predictions']
pred_df = pd.DataFrame({'pred': pred, 'pred_2': pred_2})
assert len(pred_df[pred_df['pred'] != pred_df['pred_2']]) == 0
def __init__(self, teacher: PredictiveModel,
ensembler_modeler: Modeler,
student_modelers: List = [RandomForestModeler(), AdaBoostModeler()],
model_class = Oracle
):
self.teacher = teacher
self.student_modelers = student_modelers
self.ensembler_modeler = ensembler_modeler
self.model_class = model_class
self.stats = {}
def test_ml_based_ensemble(self):
data = self.load_data()
oracle_modeler = TimeseriesOracleModeler(
teacher=RuleModel(),
student_modelers=[RandomForestModeler(),
AdaBoostModeler()],
ensembler_modeler=MyMLModeler())
num_samples = 2
oracle = oracle_modeler.build_model(
{
'unlabeled_data': data['training_data']['X'],
'labeled_data': {
'X_train': data['training_data']['X'],
'y_train': np.array(range(num_samples)),
'X_test': data['training_data']['X'],
'y_test': np.array(range(num_samples))
}
},
id_col='machineID',
ts_col='date')
pred = oracle.predict(data['training_data'])['predictions']
assert len(pred) == num_samples
with tempfile.TemporaryDirectory() as path:
os.environ['H1ST_MODEL_REPO_PATH'] = path
version = oracle.persist()
oracle_2 = TimeSeriesOracle(
teacher=RuleModel(),
students=[RandomForestModel(),
AdaBoostModel()],
ensembler=MyMLModel())
oracle_2.load_params(version)
assert 'sklearn' in str(type(oracle_2.students[0].base_model))
pred_2 = oracle_2.predict(data['training_data'])['predictions']
pred_df = pd.DataFrame({'pred': pred, 'pred_2': pred_2})
assert len(pred_df[pred_df['pred'] != pred_df['pred_2']]) == 0