def setUp(self) -> None:
self.arti = AI('test', 'test2')
data = {'col1': [1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
'col2': ['train', 'eval', 'train', 'eval', 'eval', 'train1', 'eval', 'train', 'train', 'train1']}
self.dataframe = pd.DataFrame(data=data)
def test_constructor_build(self):
data_builder = DataBuilder(data_source='C:/python/practice2/AIBuilder/tests/data/test_data.csv',
target_column='target_1',
data_columns=['feature_1', 'feature_2', 'feature_3'])
arti = AI(project_name='name', log_dir='path/to/dir')
data_builder.validate()
data_builder.build(ml_model=arti)
column_names = ['feature_1', 'feature_2', 'feature_3', 'target_1']
self.validate_data_frame(arti.training_data, column_names)
def create_AI(self, builders: list, ai_name: str = None) -> AbstractAI:
self.validate_builders(builders)
ml_model = AI(self.project_name, self.log_dir, ai_name)
self.dispatcher.dispatch(
ModelEvent(KernelDispatcher.PRE_RUN_BUILDERS, ml_model))
name = ml_model.get_name()
description = {}
for builder in builders:
self.console_printer.line('running: ' + builder.__class__.__name__)
if isinstance(builder, EstimatorBuilder):
# overwrite old name from being loaded from cache.
ml_model.set_name(name)
result = builder.build(ml_model)
if result is not None:
# Result will be None if cache prevents execution, keep using the old model as it has the description,
# which is required for determining function cache key.
# todo: fix this, we should not be passing a model just because the cache needs the description.
ml_model = result
description[builder.builder_type] = builder.describe()
ml_model.description = description
return ml_model
def run(self):
self.model = None
pre_run_event = TesterEvent(event_name=KernelDispatcher.PRE_RUN,
session=self.session,
tester=self.tester)
self.dispatcher.dispatch(pre_run_event)
while self.factory.has_next_ai():
#todo: move the new is unique logic somewhere else and changing sequence of operation seems to difficult, smell?
final_description = self.preview_model_description()
final_description_hash = AITester.stable_hash_description(
self.preview_model_description())
log_dir_path = self.factory.log_dir
if not AITester.is_hash_unique_to_report(
log_dir_path=log_dir_path,
description_hash=final_description_hash):
self.factory.builder_permutations.pop()
#todo: some kind of null model, as it is just used to pervey name and log dir?
model = AI(self.factory.project_name, self.factory.log_dir,
self.factory.project_name + '_X')
model.description = final_description
self.tester.ml_model = model
self.ModelNotUnique()
continue
self.doCreateModel()
if self.train:
self.doTrainModel()
self.all_models.append(self.model)
if self.evaluate:
self.doEvaluateModel()
if self.predict:
self.prediction_results = self.doPredict()
post_run_event = TesterEvent(event_name=KernelDispatcher.POST_RUN,
session=self.session,
tester=self.tester)
self.dispatcher.dispatch(post_run_event)
def setUp(self):
strategy_class = self.get_strategy_class_name()
self.ml_model = AI(project_name='test',
log_dir='test/dir',
name='test')
self.ml_model.optimizer = mock.Mock()
self.ml_model.training_data = mock.Mock()
self.ml_model.training_data.get_tf_feature_columns = mock.Mock()
kwargs = {'ml_model': self.ml_model}
kwargs.update(self.additional_parameters())
self.strategy = strategy_class(**kwargs)
def setUp(self):
data_array = {'feat_A': [1, 2, 3], 'feat_B': [8, 6, 4], 'target': [9, 8, 7]}
df2 = df = pd.DataFrame(data_array)
train_model = DataModel(df)
train_model.set_target_column('target')
train_model.set_feature_columns(['feat_A', 'feat_B'])
eval_model = DataModel(df2)
eval_model.set_target_column('target')
eval_model.set_feature_columns(['feat_A', 'feat_B'])
self.arti = AI('test', 'test/test')
self.arti.training_data = train_model
self.arti.evaluation_data = eval_model
def setUp(self):
self.ml_model = AI(project_name='test',
log_dir='test/dir',
name='test')
class TestCategoricalDataSplitter(unittest.TestCase):
def setUp(self) -> None:
self.arti = AI('test', 'test2')
data = {'col1': [1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
'col2': ['train', 'eval', 'train', 'eval', 'eval', 'train1', 'eval', 'train', 'train', 'train1']}
self.dataframe = pd.DataFrame(data=data)
def test_build_training_data(self):
builder = CategoricalDataSplitter(data_source='training', column_name='col2',
training_categories=['train', 'train1'])
training_model = DataModel(self.dataframe)
self.arti.set_training_data(training_model)
self.arti = builder.build(ml_model=self.arti)
split_training_data = self.arti.get_training_data().get_dataframe()
split_evaluation_data = self.arti.get_evaluation_data().get_dataframe()
self.assertEqual(6, len(split_training_data))
for item in split_training_data.values:
self.assertTrue('train' in item[1])
self.assertTrue(item[0] in [1, 2, 3, 6, 8, 9, 0], f' item {item[0]} not found.')
self.assertEqual(4, len(split_evaluation_data))
for item in split_evaluation_data.values:
self.assertEqual('eval', item[1])
self.assertTrue(item[0] in [2, 4, 5, 7], f' item {item[0]} not found.')
def test_build_evaluation_data(self):
builder = CategoricalDataSplitter(data_source='evaluation', column_name='col2',
training_categories=['train', 'train1'])
evaluation_model = DataModel(self.dataframe)
self.arti.set_evaluation_data(evaluation_model)
self.arti = builder.build(ml_model=self.arti)
split_training_data = self.arti.get_training_data().get_dataframe()
split_evaluation_data = self.arti.get_evaluation_data().get_dataframe()
self.assertEqual(6, len(split_training_data))
for item in split_training_data.values:
self.assertTrue('train' in item[1])
self.assertTrue(item[0] in [1, 2, 3, 6, 8, 9, 0], f' item {item[0]} not found.')
self.assertEqual(4, len(split_evaluation_data))
for item in split_evaluation_data.values:
self.assertEqual('eval', item[1])
self.assertTrue(item[0] in [2, 4, 5, 7], f' item {item[0]} not found.')
def test_evaluation_categories(self):
builder = CategoricalDataSplitter(data_source='training', column_name='col2',
eval_categories=['eval'])
training_model = DataModel(self.dataframe)
self.arti.set_training_data(training_model)
self.arti = builder.build(ml_model=self.arti)
split_training_data = self.arti.get_training_data().get_dataframe()
split_evaluation_data = self.arti.get_evaluation_data().get_dataframe()
self.assertEqual(6, len(split_training_data))
for item in split_training_data.values:
self.assertTrue('train' in item[1])
self.assertTrue(item[0] in [1, 2, 3, 6, 8, 9, 0], f' item {item[0]} not found.')
self.assertEqual(4, len(split_evaluation_data))
for item in split_evaluation_data.values:
self.assertEqual('eval', item[1])
self.assertTrue(item[0] in [2, 4, 5, 7], f' item {item[0]} not found.')
def test_both_categories(self):
builder = CategoricalDataSplitter(data_source='training', column_name='col2',
eval_categories=['eval'], training_categories=['train', 'train1'])
training_model = DataModel(self.dataframe)
self.arti.set_training_data(training_model)
self.arti = builder.build(ml_model=self.arti)
split_training_data = self.arti.get_training_data().get_dataframe()
split_evaluation_data = self.arti.get_evaluation_data().get_dataframe()
self.assertEqual(6, len(split_training_data))
for item in split_training_data.values:
self.assertTrue('train' in item[1])
self.assertTrue(item[0] in [1, 2, 3, 6, 8, 9, 0], f' item {item[0]} not found.')
self.assertEqual(4, len(split_evaluation_data))
for item in split_evaluation_data.values:
self.assertEqual('eval', item[1])
self.assertTrue(item[0] in [2, 4, 5, 7], f' item {item[0]} not found.')
def test_missing_categories(self):
builder = CategoricalDataSplitter(data_source='training', column_name='col2',
eval_categories=['eval'], training_categories=['train'])
training_model = DataModel(self.dataframe)
self.arti.set_training_data(training_model)
with self.assertRaises(AssertionError):
self.arti = builder.build(ml_model=self.arti)
def test_no_categories(self):
builder = CategoricalDataSplitter(data_source='training', column_name='col2')
training_model = DataModel(self.dataframe)
self.arti.set_training_data(training_model)
with self.assertRaises(AssertionError):
self.arti = builder.build(ml_model=self.arti)