def get_training_run_details(self, model_id, run_index):
model = self._client.get_model(model_id)
options = model.training_runs[run_index].training_options
details = {}
if model.training_runs[run_index].results:
details['actual_iterations'] = len(
model.training_runs[run_index].results)
for field in options.ListFields():
field_name = field[0].name
value = field[1]
if field_name == 'data_split_method':
details[field_name] = Model.DataSplitMethod(value).name
elif field_name == 'distance_type':
details[field_name] = Model.DistanceType(value).name
elif field_name == 'learn_rate_strategy':
details[field_name] = Model.LearnRateStrategy(value).name
elif field_name == 'loss_type':
details[field_name] = Model.LossType(value).name
elif field_name == 'optimization_strategy':
details[field_name] = Model.OptimizationStrategy(value).name
elif field_name == 'kmeans_initialization_method':
details[
field_name] = Model.KmeansEnums.KmeansInitializationMethod(
value).name
else:
details[field_name] = format_preview_value(value)
return {'details': details}
def testGetModelDetailsEmptyFields(self):
client = Mock()
model = Mock(project='project_id',
dataset_id='dataset_id',
model_id='model_id',
description=None,
labels={},
created=datetime.datetime(2020,
7,
14,
13,
23,
45,
67,
tzinfo=None),
expires=datetime.datetime(2021,
7,
14,
13,
23,
45,
67,
tzinfo=None),
location='US',
modified=datetime.datetime(2020,
7,
15,
15,
11,
23,
32,
tzinfo=None),
model_type=0,
label_columns=[],
feature_columns=[],
training_runs=[Mock(start_time=Mock(seconds=1234567))])
client.get_model = Mock(return_value=model)
bigquery = BigQueryService(client)
expected = {
'details': {
'id': 'project_id.dataset_id.model_id',
'name': 'model_id',
'description': None,
'labels': None,
'date_created': '2020-07-14T13:23:45.000067',
'expires': '2021-07-14T13:23:45.000067',
'location': 'US',
'last_modified': '2020-07-15T15:11:23.000032',
'model_type': Model.ModelType(0).name,
'schema_labels': [],
'feature_columns': [],
'training_runs': ['1970-01-15T06:56:07+00:00']
}
}
result = bigquery.get_model_details('some_model_id')
self.assertEqual(expected, result)
def testGetModelDetailsEmptyFields(self):
client = Mock()
model = Mock(project='project_id',
dataset_id='dataset_id',
model_id='model_id',
description=None,
labels={},
created=datetime.datetime(2020,
7,
14,
13,
23,
45,
67,
tzinfo=None),
expires=datetime.datetime(2021,
7,
14,
13,
23,
45,
67,
tzinfo=None),
location='US',
modified=datetime.datetime(2020,
7,
15,
15,
11,
23,
32,
tzinfo=None),
model_type=0,
label_columns=[],
feature_columns=[])
client.get_model = Mock(return_value=model)
expected = {
'details': {
'id': 'project_id.dataset_id.model_id',
'name': 'model_id',
'description': None,
'labels': None,
'date_created': '2020-07-14T13:23:45.000067',
'expires': '2021-07-14T13:23:45.000067',
'location': 'US',
'last_modified': '2020-07-15T15:11:23.000032',
'model_type': Model.ModelType(0).name,
'schema_labels': [],
'feature_columns': []
}
}
result = get_model_details(client, 'some_model_id')
self.assertEqual(expected, result)
def get_model_details(self, model_id):
'''Returns basic metadata for a model, as well as an array of training run dates.'''
model = self._client.get_model(model_id)
return {
'details': {
'id':
"{}.{}.{}".format(model.project, model.dataset_id,
model.model_id),
'name':
model.model_id,
'description':
model.description,
'labels': [
"{}: {}".format(label, value)
for label, value in model.labels.items()
] if model.labels else None,
'date_created':
format_date(model.created),
'last_modified':
format_date(model.modified),
'expires':
format_date(model.expires) if model.expires else None,
'location':
model.location,
'model_type':
Model.ModelType(model.model_type).name,
'schema_labels': [{
'name':
label_col.name,
'type':
SqlTypeNames[StandardSqlDataTypes(
label_col.type.type_kind).name].name
} for label_col in model.label_columns],
'feature_columns': [{
'name':
feat_col.name,
'type':
SqlTypeNames[StandardSqlDataTypes(
feat_col.type.type_kind).name].name
} for feat_col in model.feature_columns],
'training_runs': [
format_date(
datetime.datetime.fromtimestamp(
run.start_time.seconds, datetime.timezone.utc))
for run in model.training_runs
]
}
}
def get_model_details(client, model_id):
model = client.get_model(model_id)
return {
'details': {
'id':
"{}.{}.{}".format(model.project, model.dataset_id, model.model_id),
'name':
model.model_id,
'description':
model.description,
'labels': [
"{}: {}".format(label, value)
for label, value in model.labels.items()
] if model.labels else None,
'date_created':
format_date(model.created),
'last_modified':
format_date(model.modified),
'expires':
format_date(model.expires) if model.expires else None,
'location':
model.location,
'model_type':
Model.ModelType(model.model_type).name,
'schema_labels': [{
'name':
label_col.name,
'type':
SqlTypeNames[StandardSqlDataTypes(
label_col.type.type_kind).name].name
} for label_col in model.label_columns],
'feature_columns': [{
'name':
feat_col.name,
'type':
SqlTypeNames[StandardSqlDataTypes(
feat_col.type.type_kind).name].name
} for feat_col in model.feature_columns]
}
}
def testGetTrainingRunDetails(self):
client = Mock()
options = [['data_split_column', 'column_name_1'],
['data_split_eval_fraction', 0.3], ['data_split_method', 1],
['distance_type',
1], ['early_stop', BoolValue(value=False)],
['initial_learn_rate', 0.1],
['input_label_columns', ['column1', 'column2']],
['kmeans_initialization_column', 'column_name_2'],
['kmeans_initialization_method', 1],
['l1_regularization',
DoubleValue(value=0.5)],
['l2_regularization',
DoubleValue(value=0.6)],
['label_class_weights', [0.2, 0.3]], ['learn_rate', 0.7],
['learn_rate_strategy', 1], ['loss_type', 1],
['max_iterations', 20],
['min_relative_progress',
DoubleValue(value=0.4)], ['model_uri', 'model.uri.string'],
['num_clusters', 7], ['optimization_strategy', 1],
['warm_start', BoolValue(value=True)]]
mocked_options = []
for option in options:
option_mock = Mock()
option_mock.name = option[0]
mocked_options.append([option_mock, option[1]])
training_options = Mock()
training_options.ListFields = Mock(return_value=mocked_options)
training_run = Mock(training_options=training_options,
results=['result', 'result', 'result'])
model = Mock(training_runs=[training_run])
client.get_model = Mock(return_value=model)
bigquery = BigQueryService(client)
expected = {
'details': {
'actual_iterations':
3,
'data_split_column':
'column_name_1',
'data_split_eval_fraction':
0.3,
'data_split_method':
Model.DataSplitMethod(1).name,
'distance_type':
Model.DistanceType(1).name,
'early_stop':
'False',
'initial_learn_rate':
0.1,
'input_label_columns':
"['column1', 'column2']",
'kmeans_initialization_column':
'column_name_2',
'kmeans_initialization_method':
Model.KmeansEnums.KmeansInitializationMethod(1).name,
'l1_regularization':
0.5,
'l2_regularization':
0.6,
'label_class_weights':
'[0.2, 0.3]',
'learn_rate':
0.7,
'learn_rate_strategy':
Model.LearnRateStrategy(1).name,
'loss_type':
Model.LossType(1).name,
'max_iterations':
20,
'min_relative_progress':
0.4,
'model_uri':
'model.uri.string',
'num_clusters':
7,
'optimization_strategy':
Model.OptimizationStrategy(1).name,
'warm_start':
'True'
}
}
result = bigquery.get_training_run_details('some_model_id', 0)
self.assertEqual(expected, result)
def testGetModelDetailsFull(self):
client = Mock()
label_col_0 = Mock()
label_col_0.name = 'schema_label_0'
label_col_0.type = Mock(type_kind=7)
label_col_1 = Mock()
label_col_1.name = 'schema_label_1'
label_col_1.type = Mock(type_kind=9)
feature_col_0 = Mock()
feature_col_0.name = 'feature_col_0'
feature_col_0.type = Mock(type_kind=8)
model = Mock(project='project_id',
dataset_id='dataset_id',
model_id='model_id',
description='description of model',
labels={
'label_0': 'value_0',
'label_1': 'value_1'
},
created=datetime.datetime(2020,
7,
14,
13,
23,
45,
67,
tzinfo=None),
expires=datetime.datetime(2021,
7,
14,
13,
23,
45,
67,
tzinfo=None),
location='US',
modified=datetime.datetime(2020,
7,
15,
15,
11,
23,
32,
tzinfo=None),
model_type=0,
label_columns=[label_col_0, label_col_1],
feature_columns=[feature_col_0],
training_runs=[
Mock(start_time=Mock(seconds=1234567)),
Mock(start_time=Mock(seconds=1234789))
])
client.get_model = Mock(return_value=model)
bigquery = BigQueryService(client)
expected = {
'details': {
'id':
'project_id.dataset_id.model_id',
'name':
'model_id',
'description':
'description of model',
'labels': ['label_0: value_0', 'label_1: value_1'],
'date_created':
'2020-07-14T13:23:45.000067',
'last_modified':
'2020-07-15T15:11:23.000032',
'expires':
'2021-07-14T13:23:45.000067',
'location':
'US',
'model_type':
Model.ModelType(0).name,
'schema_labels': [{
'name':
'schema_label_0',
'type':
SqlTypeNames[StandardSqlDataTypes(7).name].name
}, {
'name':
'schema_label_1',
'type':
SqlTypeNames[StandardSqlDataTypes(9).name].name
}],
'feature_columns': [{
'name':
'feature_col_0',
'type':
SqlTypeNames[StandardSqlDataTypes(8).name].name
}],
'training_runs':
['1970-01-15T06:56:07+00:00', '1970-01-15T06:59:49+00:00']
}
}
result = bigquery.get_model_details('some_model_id')
self.assertEqual(expected, result)