def test_predictor_save_predictions(matrix_type, predict_setup_args):
(project_storage, db_engine, model_id) = predict_setup_args
# if save_predictions is sent as False, don't save
predictor = Predictor(project_storage.model_storage_engine(),
db_engine,
save_predictions=False)
matrix = matrix_creator(index="entity_id")
metadata = matrix_metadata_creator(end_time=AS_OF_DATE,
matrix_type=matrix_type,
indices=["entity_id"])
matrix_store = get_matrix_store(project_storage, matrix, metadata)
train_matrix_columns = matrix.columns[0:-1].tolist()
predict_proba = predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=train_matrix_columns,
)
# assert
# 1. that the returned predictions are of the desired length
assert len(predict_proba) == 2
# 2. that the predictions table entries are present and
# can be linked to the original models
assert not table_has_data(f"{matrix_type}_predictions", db_engine)
def test_predictor_get_train_columns():
with prepare() as (project_storage, db_engine, model_id):
predictor = Predictor(project_storage.model_storage_engine(),
db_engine)
train_store = get_matrix_store(
project_storage=project_storage,
matrix=matrix_creator(),
metadata=matrix_metadata_creator(matrix_type="train"),
)
# flip the order of some feature columns in the test matrix
other_order_matrix = matrix_creator()
order = other_order_matrix.columns.tolist()
order[0], order[1] = order[1], order[0]
other_order_matrix = other_order_matrix[order]
test_store = get_matrix_store(
project_storage=project_storage,
matrix=other_order_matrix,
metadata=matrix_metadata_creator(matrix_type="test"),
)
# Runs the same test for training and testing predictions
for store, mat_type in zip((train_store, test_store),
("train", "test")):
predict_proba = predictor.predict(
model_id,
store,
misc_db_parameters=dict(),
train_matrix_columns=train_store.columns(),
)
# assert
# 1. that we calculated predictions
assert len(predict_proba) > 0
# 2. that the predictions table entries are present and
# can be linked to the original models
records = [
row
for row in db_engine.execute("""select entity_id, as_of_date
from {}_results.predictions
join model_metadata.models using (model_id)""".format(
mat_type, mat_type))
]
assert len(records) > 0
def test_predictor_retrieve():
with prepare() as (project_storage, db_engine, model_id):
predictor = Predictor(project_storage.model_storage_engine(),
db_engine,
replace=False)
# create prediction set
matrix = matrix_creator()
metadata = matrix_metadata_creator()
matrix_store = get_matrix_store(project_storage, matrix, metadata)
predict_proba = predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=matrix.columns[0:-1].tolist())
# When run again, the predictions retrieved from the database
# should match.
#
# Some trickiness here. Let's explain:
#
# If we are not careful, retrieving predictions from the database and
# presenting them as a numpy array can result in a bad ordering,
# since the given matrix may not be 'ordered' by some criteria
# that can be easily represented by an ORDER BY clause.
#
# It will sometimes work, because without ORDER BY you will get
# it back in the table's physical order, which unless something has
# happened to the table will be the order you inserted it,
# which could very well be the order in the matrix.
# So it's not a bug that would necessarily immediately show itself,
# but when it does go wrong your scores will be garbage.
#
# So we simulate a table order mutation that can happen over time:
# Remove the first row and put it at the end.
# If the Predictor doesn't explicitly reorder the results, this will fail
# Only running on TestPrediction because TrainPrediction behaves the exact same way
reorder_session = sessionmaker(bind=db_engine)()
obj = reorder_session.query(TestPrediction).first()
reorder_session.delete(obj)
reorder_session.commit()
make_transient(obj)
reorder_session = sessionmaker(bind=db_engine)()
reorder_session.add(obj)
reorder_session.commit()
predictor.load_model = Mock()
new_predict_proba = predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=matrix.columns[0:-1].tolist())
assert_array_equal(new_predict_proba, predict_proba)
assert not predictor.load_model.called
def test_calculate_and_save():
with rig_engines() as (db_engine, project_storage):
train_store = get_matrix_store(
project_storage,
matrix_creator(),
matrix_metadata_creator(matrix_type='train'),
)
test_store = get_matrix_store(
project_storage,
matrix_creator(),
matrix_metadata_creator(matrix_type='test'),
)
calculator = IndividualImportanceCalculator(db_engine,
methods=['sample'],
replace=False)
# given a trained model
# and a test matrix
_, model_id = \
fake_trained_model(
db_engine,
train_matrix_uuid=train_store.uuid
)
# i expect to be able to call calculate and save
calculator.calculate_and_save_all_methods_and_dates(
model_id, test_store)
# and find individual importances in the results schema afterwards
records = [
row for row in db_engine.execute('''select entity_id, as_of_date
from test_results.individual_importances
join model_metadata.models using (model_id)''')
]
assert len(records) > 0
# and that when run again, has the same result
calculator.calculate_and_save_all_methods_and_dates(
model_id, test_store)
new_records = [
row for row in db_engine.execute('''select entity_id, as_of_date
from test_results.individual_importances
join model_metadata.models using (model_id)''')
]
assert len(records) == len(new_records)
assert records == new_records
def test_predictor_needs_predictions(matrix_type, predict_setup_args):
(project_storage, db_engine, model_id) = predict_setup_args
# if not all of the predictions for the given model id and matrix are present in the db,
# needs_predictions should return true. else, false
predictor = Predictor(project_storage.model_storage_engine(), db_engine)
matrix = matrix_creator(index="entity_id")
metadata = matrix_metadata_creator(end_time=AS_OF_DATE,
matrix_type=matrix_type,
indices=["entity_id"])
matrix_store = get_matrix_store(project_storage, matrix, metadata)
train_matrix_columns = matrix.columns[0:-1].tolist()
# we haven't done anything yet, this should definitely need predictions
assert predictor.needs_predictions(matrix_store, model_id)
predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=train_matrix_columns,
)
# now that predictions have been made, this should no longer need predictions
assert not predictor.needs_predictions(matrix_store, model_id)
def test_integration():
with rig_engines() as (db_engine, project_storage):
train_store = get_matrix_store(
project_storage, matrix_creator(),
matrix_metadata_creator(matrix_type='train'))
as_of_dates = [datetime.date(2016, 12, 21), datetime.date(2017, 1, 21)]
test_stores = []
for as_of_date in as_of_dates:
matrix_store = get_matrix_store(
project_storage,
pandas.DataFrame.from_dict({
'entity_id': [3],
'feature_one': [8],
'feature_two': [5],
'label': [0]
}).set_index('entity_id'),
matrix_metadata_creator(end_time=as_of_date,
indices=['entity_id']))
test_stores.append(matrix_store)
model_storage_engine = ModelStorageEngine(project_storage)
experiment_hash = save_experiment_and_get_hash({}, db_engine)
# instantiate pipeline objects
trainer = ModelTrainer(
experiment_hash=experiment_hash,
model_storage_engine=model_storage_engine,
db_engine=db_engine,
)
predictor = Predictor(model_storage_engine, db_engine)
model_evaluator = ModelEvaluator([{
'metrics': ['precision@'],
'thresholds': {
'top_n': [5]
}
}], [{}], db_engine)
# run the pipeline
grid_config = {
'sklearn.linear_model.LogisticRegression': {
'C': [0.00001, 0.0001],
'penalty': ['l1', 'l2'],
'random_state': [2193]
}
}
model_ids = trainer.train_models(grid_config=grid_config,
misc_db_parameters=dict(),
matrix_store=train_store)
for model_id in model_ids:
for as_of_date, test_store in zip(as_of_dates, test_stores):
predictions_proba = predictor.predict(
model_id,
test_store,
misc_db_parameters=dict(),
train_matrix_columns=['feature_one', 'feature_two'])
model_evaluator.evaluate(
predictions_proba,
test_store,
model_id,
)
# assert
# 1. that the predictions table entries are present and
# can be linked to the original models
records = [
row for row in db_engine.execute(
'''select entity_id, model_id, as_of_date
from test_results.predictions
join model_metadata.models using (model_id)
order by 3, 2''')
]
assert records == [
(3, 1, datetime.datetime(2016, 12, 21)),
(3, 2, datetime.datetime(2016, 12, 21)),
(3, 3, datetime.datetime(2016, 12, 21)),
(3, 4, datetime.datetime(2016, 12, 21)),
(3, 1, datetime.datetime(2017, 1, 21)),
(3, 2, datetime.datetime(2017, 1, 21)),
(3, 3, datetime.datetime(2017, 1, 21)),
(3, 4, datetime.datetime(2017, 1, 21)),
]
# that evaluations are there
records = [
row for row in db_engine.execute('''
select model_id, evaluation_start_time, metric, parameter
from test_results.evaluations order by 2, 1''')
]
assert records == [
(1, datetime.datetime(2016, 12, 21), 'precision@', '5_abs'),
(2, datetime.datetime(2016, 12, 21), 'precision@', '5_abs'),
(3, datetime.datetime(2016, 12, 21), 'precision@', '5_abs'),
(4, datetime.datetime(2016, 12, 21), 'precision@', '5_abs'),
(1, datetime.datetime(2017, 1, 21), 'precision@', '5_abs'),
(2, datetime.datetime(2017, 1, 21), 'precision@', '5_abs'),
(3, datetime.datetime(2017, 1, 21), 'precision@', '5_abs'),
(4, datetime.datetime(2017, 1, 21), 'precision@', '5_abs'),
]
def test_predictor_entity_index():
with prepare() as (project_storage, db_engine, model_id):
predictor = Predictor(project_storage.model_storage_engine(),
db_engine)
# Runs the same test for training and testing predictions
for mat_type in ("train", "test"):
matrix = matrix_creator(index="entity_id")
metadata = matrix_metadata_creator(end_time=AS_OF_DATE,
matrix_type=mat_type,
indices=["entity_id"])
matrix_store = get_matrix_store(project_storage, matrix, metadata)
train_matrix_columns = matrix.columns[0:-1].tolist()
predict_proba = predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=train_matrix_columns,
)
# assert
# 1. that the returned predictions are of the desired length
assert len(predict_proba) == 2
# 2. that the predictions table entries are present and
# can be linked to the original models
records = [
row
for row in db_engine.execute("""select entity_id, as_of_date
from {}_results.predictions
join model_metadata.models using (model_id)""".format(
mat_type, mat_type))
]
assert len(records) == 2
# 3. that the contained as_of_dates match what we sent in
for record in records:
assert record[1].date() == AS_OF_DATE
# 4. that the entity ids match the given dataset
assert sorted([record[0] for record in records]) == [1, 2]
# 5. running with same model_id, different as of date
# then with same as of date only replaces the records
# with the same date
# Runs the same test for training and testing predictions
for mat_type in ("train", "test"):
new_matrix = matrix_creator(index="entity_id")
new_metadata = matrix_metadata_creator(
end_time=AS_OF_DATE + datetime.timedelta(days=1),
matrix_type=mat_type,
indices=["entity_id"],
)
new_matrix_store = get_matrix_store(project_storage, new_matrix,
new_metadata)
predictor.predict(
model_id,
new_matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=train_matrix_columns,
)
predictor.predict(
model_id,
matrix_store,
misc_db_parameters=dict(),
train_matrix_columns=train_matrix_columns,
)
records = [
row
for row in db_engine.execute("""select entity_id, as_of_date
from {}_results.predictions
join model_metadata.models using (model_id)""".format(
mat_type, mat_type))
]
assert len(records) == 4
# 6. That we can delete the model when done prediction on it
predictor.delete_model(model_id)
assert predictor.load_model(model_id) is None
def test_integration():
with rig_engines() as (db_engine, project_storage):
train_store = get_matrix_store(
project_storage,
matrix_creator(),
matrix_metadata_creator(matrix_type="train"),
)
as_of_dates = [datetime.date(2016, 12, 21), datetime.date(2017, 1, 21)]
test_stores = []
for as_of_date in as_of_dates:
matrix_store = get_matrix_store(
project_storage,
pandas.DataFrame.from_dict({
"entity_id": [3],
"feature_one": [8],
"feature_two": [5],
"label": [0],
}).set_index("entity_id"),
matrix_metadata_creator(end_time=as_of_date,
indices=["entity_id"]),
)
test_stores.append(matrix_store)
model_storage_engine = ModelStorageEngine(project_storage)
experiment_hash = save_experiment_and_get_hash({}, db_engine)
# instantiate pipeline objects
trainer = ModelTrainer(
experiment_hash=experiment_hash,
model_storage_engine=model_storage_engine,
db_engine=db_engine,
)
predictor = Predictor(model_storage_engine, db_engine)
model_evaluator = ModelEvaluator([{
"metrics": ["precision@"],
"thresholds": {
"top_n": [5]
}
}], [{}], db_engine)
# run the pipeline
grid_config = {
"sklearn.linear_model.LogisticRegression": {
"C": [0.00001, 0.0001],
"penalty": ["l1", "l2"],
"random_state": [2193],
}
}
model_ids = trainer.train_models(grid_config=grid_config,
misc_db_parameters=dict(),
matrix_store=train_store)
for model_id in model_ids:
for as_of_date, test_store in zip(as_of_dates, test_stores):
predictions_proba = predictor.predict(
model_id,
test_store,
misc_db_parameters=dict(),
train_matrix_columns=["feature_one", "feature_two"],
)
model_evaluator.evaluate(predictions_proba, test_store,
model_id)
# assert
# 1. that the predictions table entries are present and
# can be linked to the original models
records = [
row for row in db_engine.execute(
"""select entity_id, model_id, as_of_date
from test_results.predictions
join model_metadata.models using (model_id)
order by 3, 2""")
]
assert records == [
(3, 1, datetime.datetime(2016, 12, 21)),
(3, 2, datetime.datetime(2016, 12, 21)),
(3, 3, datetime.datetime(2016, 12, 21)),
(3, 4, datetime.datetime(2016, 12, 21)),
(3, 1, datetime.datetime(2017, 1, 21)),
(3, 2, datetime.datetime(2017, 1, 21)),
(3, 3, datetime.datetime(2017, 1, 21)),
(3, 4, datetime.datetime(2017, 1, 21)),
]
# that evaluations are there
records = [
row for row in db_engine.execute("""
select model_id, evaluation_start_time, metric, parameter
from test_results.evaluations order by 2, 1""")
]
assert records == [
(1, datetime.datetime(2016, 12, 21), "precision@", "5_abs"),
(2, datetime.datetime(2016, 12, 21), "precision@", "5_abs"),
(3, datetime.datetime(2016, 12, 21), "precision@", "5_abs"),
(4, datetime.datetime(2016, 12, 21), "precision@", "5_abs"),
(1, datetime.datetime(2017, 1, 21), "precision@", "5_abs"),
(2, datetime.datetime(2017, 1, 21), "precision@", "5_abs"),
(3, datetime.datetime(2017, 1, 21), "precision@", "5_abs"),
(4, datetime.datetime(2017, 1, 21), "precision@", "5_abs"),
]