def test_initialize_models():
test_tablename, _ = create_dha(path='data/dha_missing.csv')
engine = Engine(seed=0)
num_models = 5
engine.initialize_models(test_tablename, num_models)
model_ids = engine.persistence_layer.get_model_ids(test_tablename)
assert sorted(model_ids) == range(num_models)
for i in range(num_models):
model = engine.persistence_layer.get_models(test_tablename, i)
assert model['iterations'] == 0
def test_initialize_models():
test_tablename, _ = create_dha(path='data/dha_missing.csv')
engine = Engine(seed=0)
num_models = 5
engine.initialize_models(test_tablename, num_models)
model_ids = engine.persistence_layer.get_model_ids(test_tablename)
assert sorted(model_ids) == range(num_models)
for i in range(num_models):
model = engine.persistence_layer.get_models(test_tablename, i)
assert model['iterations'] == 0
def test_infer():
# TODO: whereclauses
test_tablename, _ = create_dha(path='data/dha_missing.csv')
# dha_missing has missing qual_score in first 5 rows, and missing name in rows 6-10.
engine = Engine(seed=0)
engine.initialize_models(test_tablename, 20)
functions = bql.bql_statement.parseString('infer name, qual_score from test',
parseAll=True).functions
whereclause = None
limit = float('inf')
order_by = False
numsamples = 30
confidence = 0
infer_result = engine.infer(test_tablename, functions, confidence, whereclause, limit,
numsamples, order_by)
assert 'column_labels' in infer_result
assert 'data' in infer_result
assert infer_result['column_labels'] == ['key', 'name', 'qual_score']
# 307 is the total number of rows in the dataset.
assert(len(infer_result['data']) == 307 and
len(infer_result['data'][0]) == len(infer_result['column_labels']))
assert type(infer_result['data'][0][0]) == numpy.string_ # type of key is int
t = type(infer_result['data'][0][1])
assert (t == unicode) or (t == numpy.string_) # type of name is string
assert type(infer_result['data'][0][2]) == float # type of qual_score is float
all_possible_names = [infer_result['data'][row][1] for row in range(5) + range(10, 307)]
all_observed_qual_scores = [infer_result['data'][row][2] for row in range(5, 307)]
for row in range(5):
inferred_name = infer_result['data'][row+5][1]
inferred_qual_score = infer_result['data'][row][2]
assert inferred_name in all_possible_names
assert type(inferred_qual_score) == type(1.2)
assert inferred_qual_score > min(all_observed_qual_scores)
assert inferred_qual_score < max(all_observed_qual_scores)
# Now, try infer with higher confidence, and make sure that name isn't inferred anymore.
confidence = 0.9
infer_result = engine.infer(test_tablename, functions, confidence, whereclause, limit,
numsamples, order_by)
for row in range(5):
# TODO: what do missing values look like? these should be missing
inferred_name = infer_result['data'][row+5][1]
inferred_qual_score = infer_result['data'][row][2]
assert numpy.isnan(inferred_name)
assert numpy.isnan(inferred_qual_score)
def test_infer():
# TODO: whereclauses
test_tablename, _ = create_dha(path='data/dha_missing.csv')
# dha_missing has missing qual_score in first 5 rows, and missing name in rows 6-10.
engine = Engine(seed=0)
engine.initialize_models(test_tablename, 20)
functions = bql.bql_statement.parseString('infer name, qual_score from test',
parseAll=True).functions
whereclause = None
limit = float('inf')
order_by = False
numsamples = 30
confidence = 0
infer_result = engine.infer(test_tablename, functions, confidence, whereclause, limit,
numsamples, order_by)
assert 'column_labels' in infer_result
assert 'data' in infer_result
assert infer_result['column_labels'] == ['key', 'name', 'qual_score']
# 307 is the total number of rows in the dataset.
assert(len(infer_result['data']) == 307 and
len(infer_result['data'][0]) == len(infer_result['column_labels']))
assert type(infer_result['data'][0][0]) == numpy.string_ # type of key is int
t = type(infer_result['data'][0][1])
assert (t == unicode) or (t == numpy.string_) # type of name is string
assert type(infer_result['data'][0][2]) == float # type of qual_score is float
all_possible_names = [infer_result['data'][row][1] for row in range(5) + range(10, 307)]
all_observed_qual_scores = [infer_result['data'][row][2] for row in range(5, 307)]
for row in range(5):
inferred_name = infer_result['data'][row+5][1]
inferred_qual_score = infer_result['data'][row][2]
assert inferred_name in all_possible_names
assert type(inferred_qual_score) == type(1.2)
assert inferred_qual_score > min(all_observed_qual_scores)
assert inferred_qual_score < max(all_observed_qual_scores)
# Now, try infer with higher confidence, and make sure that name isn't inferred anymore.
confidence = 0.9
infer_result = engine.infer(test_tablename, functions, confidence, whereclause, limit,
numsamples, order_by)
for row in range(5):
# TODO: what do missing values look like? these should be missing
inferred_name = infer_result['data'][row+5][1]
inferred_qual_score = infer_result['data'][row][2]
assert numpy.isnan(inferred_name)
assert numpy.isnan(inferred_qual_score)
