def test_dfs_empty_features():
error_text = "No features can be generated from the specified primitives. Please make sure the primitives you are using are compatible with the variable types in your data."
teams = pd.DataFrame({"id": range(3), "name": ["Breakers", "Spirit", "Thorns"]})
games = pd.DataFrame(
{
"id": range(5),
"home_team_id": [2, 2, 1, 0, 1],
"away_team_id": [1, 0, 2, 1, 0],
"home_team_score": [3, 0, 1, 0, 4],
"away_team_score": [2, 1, 2, 0, 0],
}
)
dataframes = {
"teams": (teams, "id", None, {"name": "natural_language"}),
"games": (games, "id"),
}
relationships = [("teams", "id", "games", "home_team_id")]
with patch.object(DeepFeatureSynthesis, "build_features", return_value=[]):
features = dfs(
dataframes, relationships, target_dataframe_name="teams", features_only=True
)
assert features == []
with pytest.raises(AssertionError, match=error_text), patch.object(
DeepFeatureSynthesis, "build_features", return_value=[]
):
dfs(
dataframes,
relationships,
target_dataframe_name="teams",
features_only=False,
)
def test_no_warns_with_camel_and_title_case(es):
for trans_primitive in ["isNull", "IsNull"]:
# Should not raise a UnusedPrimitiveWarning warning
with pytest.warns(None) as record:
dfs(
entityset=es,
target_dataframe_name="customers",
trans_primitives=[trans_primitive],
max_depth=1,
features_only=True,
)
assert not record
for agg_primitive in ["numUnique", "NumUnique"]:
# Should not raise a UnusedPrimitiveWarning warning
with pytest.warns(None) as record:
dfs(
entityset=es,
target_dataframe_name="customers",
agg_primitives=[agg_primitive],
max_depth=2,
features_only=True,
)
assert not record
def test_warns_with_unused_primitives(es):
if es.dataframe_type == Library.KOALAS.value:
pytest.skip('Koalas throws extra warnings')
trans_primitives = ['num_characters', 'num_words', 'add_numeric']
agg_primitives = [Max, 'min']
warning_text = "Some specified primitives were not used during DFS:\n" + \
" trans_primitives: ['add_numeric']\n agg_primitives: ['max', 'min']\n" + \
"This may be caused by a using a value of max_depth that is too small, not setting interesting values, " + \
"or it may indicate no compatible columns for the primitive were found in the data."
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(entityset=es,
target_dataframe_name='customers',
trans_primitives=trans_primitives,
agg_primitives=agg_primitives,
max_depth=1,
features_only=True)
assert record[0].message.args[0] == warning_text
# Should not raise a warning
with pytest.warns(None) as record:
dfs(entityset=es,
target_dataframe_name='customers',
trans_primitives=trans_primitives,
agg_primitives=agg_primitives,
max_depth=2,
features_only=True)
assert not record
def test_calls_progress_callback(entities, relationships):
class MockProgressCallback:
def __init__(self):
self.total_update = 0
self.total_progress_percent = 0
def __call__(self, update, progress_percent, time_elapsed):
self.total_update += update
self.total_progress_percent = progress_percent
mock_progress_callback = MockProgressCallback()
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
progress_callback=mock_progress_callback)
assert np.isclose(mock_progress_callback.total_update, 100.0)
assert np.isclose(mock_progress_callback.total_progress_percent, 100.0)
# test with multiple jobs
mock_progress_callback = MockProgressCallback()
with cluster() as (scheduler, [a, b]):
dkwargs = {'cluster': scheduler['address']}
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
progress_callback=mock_progress_callback,
dask_kwargs=dkwargs)
assert np.isclose(mock_progress_callback.total_update, 100.0)
assert np.isclose(mock_progress_callback.total_progress_percent, 100.0)
def test_calls_progress_callback_cluster(pd_dataframes, relationships, dask_cluster):
class MockProgressCallback:
def __init__(self):
self.progress_history = []
self.total_update = 0
self.total_progress_percent = 0
def __call__(self, update, progress_percent, time_elapsed):
self.total_update += update
self.total_progress_percent = progress_percent
self.progress_history.append(progress_percent)
mock_progress_callback = MockProgressCallback()
dkwargs = {"cluster": dask_cluster.scheduler.address}
dfs(
dataframes=pd_dataframes,
relationships=relationships,
target_dataframe_name="transactions",
progress_callback=mock_progress_callback,
dask_kwargs=dkwargs,
)
assert np.isclose(mock_progress_callback.total_update, 100.0)
assert np.isclose(mock_progress_callback.total_progress_percent, 100.0)
def test_calls_progress_callback(dataframes, relationships):
class MockProgressCallback:
def __init__(self):
self.progress_history = []
self.total_update = 0
self.total_progress_percent = 0
def __call__(self, update, progress_percent, time_elapsed):
self.total_update += update
self.total_progress_percent = progress_percent
self.progress_history.append(progress_percent)
mock_progress_callback = MockProgressCallback()
dfs(
dataframes=dataframes,
relationships=relationships,
target_dataframe_name="transactions",
progress_callback=mock_progress_callback,
)
# second to last entry is the last update from feature calculation
assert np.isclose(
mock_progress_callback.progress_history[-2],
FEATURE_CALCULATION_PERCENTAGE * 100,
)
assert np.isclose(mock_progress_callback.total_update, 100.0)
assert np.isclose(mock_progress_callback.total_progress_percent, 100.0)
def test_warns_with_unused_groupby_primitives(pd_es):
warning_text = (
"Some specified primitives were not used during DFS:\n"
+ " groupby_trans_primitives: ['cum_sum']\n"
+ "This may be caused by a using a value of max_depth that is too small, not setting interesting values, "
+ "or it may indicate no compatible columns for the primitive were found in the data. If the DFS call "
+ "contained multiple instances of a primitive in the list above, none of them were used."
)
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(
entityset=pd_es,
target_dataframe_name="sessions",
groupby_trans_primitives=["cum_sum"],
max_depth=1,
features_only=True,
)
assert record[0].message.args[0] == warning_text
# Should not raise a warning
with pytest.warns(None) as record:
dfs(
entityset=pd_es,
target_dataframe_name="customers",
groupby_trans_primitives=["cum_sum"],
max_depth=1,
features_only=True,
)
assert not record
def test_warns_with_unused_primitives(es):
if ks and any(isinstance(e.df, ks.DataFrame) for e in es.entities):
pytest.skip('Koalas throws extra warnings')
trans_primitives = ['num_characters', 'num_words', 'add_numeric']
agg_primitives = [Max, 'min']
warning_text = "Some specified primitives were not used during DFS:\n" + \
" trans_primitives: ['add_numeric']\n agg_primitives: ['max', 'min']\n" + \
"This may be caused by a using a value of max_depth that is too small, not setting interesting values, " + \
"or it may indicate no compatible variable types for the primitive were found in the data."
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(entityset=es,
target_entity='customers',
trans_primitives=trans_primitives,
agg_primitives=agg_primitives,
max_depth=1)
assert record[0].message.args[0] == warning_text
# Should not raise a warning
with pytest.warns(None) as record:
dfs(entityset=es,
target_entity='customers',
trans_primitives=trans_primitives,
agg_primitives=agg_primitives,
max_depth=2)
assert not record
def test_dask_kwargs(pd_dataframes, relationships, dask_cluster):
cutoff_times_df = pd.DataFrame({
"instance_id": [1, 2, 3],
"time": [10, 12, 15]
})
feature_matrix, features = dfs(
dataframes=pd_dataframes,
relationships=relationships,
target_dataframe_name="transactions",
cutoff_time=cutoff_times_df,
)
dask_kwargs = {"cluster": dask_cluster.scheduler.address}
feature_matrix_2, features_2 = dfs(
dataframes=pd_dataframes,
relationships=relationships,
target_dataframe_name="transactions",
cutoff_time=cutoff_times_df,
dask_kwargs=dask_kwargs,
)
assert all(f1.unique_name() == f2.unique_name()
for f1, f2 in zip(features, features_2))
for column in feature_matrix:
for x, y in zip(feature_matrix[column], feature_matrix_2[column]):
assert (pd.isnull(x) and pd.isnull(y)) or (x == y)
def test_accepts_relative_training_window(datetime_es):
feature_matrix, features = dfs(entityset=datetime_es,
target_entity="transactions")
feature_matrix_2, features_2 = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-4-1 04:00"))
feature_matrix_3, features_3 = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-4-1 04:00"),
training_window=Timedelta("3 months"))
feature_matrix_4, features_4 = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-4-1 04:00"),
training_window="3 months")
# Test case for leap years
feature_matrix_5, features_5 = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-2-29 04:00"),
training_window=Timedelta("1 year"))
assert (feature_matrix.index == [1, 2, 3, 4, 5]).all()
assert (feature_matrix_2.index == [1, 2, 3, 4]).all()
assert (feature_matrix_3.index == [2, 3, 4]).all()
assert (feature_matrix_4.index == [2, 3, 4]).all()
assert (feature_matrix_5.index == [1, 2]).all()
def fit(self, cuttof_time_ids, y=None):
"""Wrapper for DFS
Calculates a feature matrix and features given a dictionary of
entities and a list of relationships.
Args:
cuttof_time_ids (list | DataFrame): Instances filtered to
calculate features on.
See Also:
:func:`synthesis.dfs`
"""
if isinstance(cuttof_time_ids, (list, np.ndarray, pd.Series)):
self.feature_defs = dfs(entities=self.entities,
relationships=self.relationships,
entityset=self.entityset,
target_entity=self.target_entity,
instance_ids=cuttof_time_ids,
agg_primitives=self.agg_primitives,
trans_primitives=self.trans_primitives,
allowed_paths=self.allowed_paths,
max_depth=self.max_depth,
ignore_entities=self.ignore_entities,
ignore_variables=self.ignore_variables,
seed_features=self.seed_features,
drop_contains=self.drop_contains,
drop_exact=self.drop_exact,
where_primitives=self.where_primitives,
max_features=self.max_features,
features_only=True,
verbose=self.verbose)
elif isinstance(cuttof_time_ids, pd.DataFrame):
self.feature_defs = dfs(entities=self.entities,
relationships=self.relationships,
entityset=self.entityset,
target_entity=self.target_entity,
cutoff_time=cuttof_time_ids,
agg_primitives=self.agg_primitives,
trans_primitives=self.trans_primitives,
allowed_paths=self.allowed_paths,
max_depth=self.max_depth,
ignore_entities=self.ignore_entities,
ignore_variables=self.ignore_variables,
seed_features=self.seed_features,
drop_contains=self.drop_contains,
drop_exact=self.drop_exact,
where_primitives=self.where_primitives,
max_features=self.max_features,
features_only=True,
verbose=self.verbose)
else:
raise TypeError(
'instance_ids must be a list, np.ndarray, pd.Series, or pd.DataFrame'
)
return self
def fit(self, cuttof_time_ids, y=None):
"""Wrapper for DFS
Calculates a feature matrix and features given a dictionary of
entities and a list of relationships.
Args:
cuttof_time_ids (list | DataFrame): Instances filtered to
calculate features on.
See Also:
:func:`synthesis.dfs`
"""
if isinstance(cuttof_time_ids, (list, np.ndarray, pd.Series)):
self.feature_defs = dfs(entities=self.entities,
relationships=self.relationships,
entityset=self.entityset,
target_entity=self.target_entity,
instance_ids=cuttof_time_ids,
agg_primitives=self.agg_primitives,
trans_primitives=self.trans_primitives,
allowed_paths=self.allowed_paths,
max_depth=self.max_depth,
ignore_entities=self.ignore_entities,
ignore_variables=self.ignore_variables,
seed_features=self.seed_features,
drop_contains=self.drop_contains,
drop_exact=self.drop_exact,
where_primitives=self.where_primitives,
max_features=self.max_features,
features_only=True,
verbose=self.verbose)
elif isinstance(cuttof_time_ids, pd.DataFrame):
self.feature_defs = dfs(entities=self.entities,
relationships=self.relationships,
entityset=self.entityset,
target_entity=self.target_entity,
cutoff_time=cuttof_time_ids,
agg_primitives=self.agg_primitives,
trans_primitives=self.trans_primitives,
allowed_paths=self.allowed_paths,
max_depth=self.max_depth,
ignore_entities=self.ignore_entities,
ignore_variables=self.ignore_variables,
seed_features=self.seed_features,
drop_contains=self.drop_contains,
drop_exact=self.drop_exact,
where_primitives=self.where_primitives,
max_features=self.max_features,
features_only=True,
verbose=self.verbose)
else:
raise TypeError('instance_ids must be a list, np.ndarray, pd.Series, or pd.DataFrame')
return self
def test_groupby_with_multioutput_primitive(pd_es):
class MultiCumSum(TransformPrimitive):
name = "multi_cum_sum"
input_types = [ColumnSchema(semantic_tags={"numeric"})]
return_type = ColumnSchema(semantic_tags={"numeric"})
number_output_features = 3
def get_function(self):
def multi_cum_sum(x):
return x.cumsum(), x.cummax(), x.cummin()
return multi_cum_sum
fm, _ = dfs(
entityset=pd_es,
target_dataframe_name="customers",
trans_primitives=[],
agg_primitives=[],
groupby_trans_primitives=[MultiCumSum, CumSum, CumMax, CumMin],
)
# Calculate output in a separate DFS call to make sure the multi-output code
# does not alter any values
fm2, _ = dfs(
entityset=pd_es,
target_dataframe_name="customers",
trans_primitives=[],
agg_primitives=[],
groupby_trans_primitives=[CumSum, CumMax, CumMin],
)
answer_cols = [
["CUM_SUM(age) by cohort", "CUM_SUM(age) by région_id"],
["CUM_MAX(age) by cohort", "CUM_MAX(age) by région_id"],
["CUM_MIN(age) by cohort", "CUM_MIN(age) by région_id"],
]
for i in range(3):
# Check that multi-output gives correct answers
f = "MULTI_CUM_SUM(age)[%d] by cohort" % i
assert f in fm.columns
for x, y in zip(fm[f].values, fm[answer_cols[i][0]].values):
assert x == y
f = "MULTI_CUM_SUM(age)[%d] by région_id" % i
assert f in fm.columns
for x, y in zip(fm[f].values, fm[answer_cols[i][1]].values):
assert x == y
# Verify single output results are unchanged by inclusion of
# multi-output primitive
for x, y in zip(fm[answer_cols[i][0]], fm2[answer_cols[i][0]]):
assert x == y
for x, y in zip(fm[answer_cols[i][1]], fm2[answer_cols[i][1]]):
assert x == y
def test_warns_cutoff_time_dask(dataframes, relationships):
cutoff_times_df = pd.DataFrame({
"instance_id": [1, 2, 3],
"time": [10, 12, 15]
})
cutoff_times_df = dd.from_pandas(cutoff_times_df, npartitions=2)
match = "cutoff_time should be a Pandas DataFrame: " \
"computing cutoff_time, this may take a while"
with pytest.warns(UserWarning, match=match):
dfs(dataframes=dataframes,
relationships=relationships,
target_dataframe_name="transactions",
cutoff_time=cutoff_times_df)
def test_accepts_pd_dateoffset_training_window(datetime_es):
# TODO: Update to use Dask dataframes when issue #882 is closed
feature_matrix, _ = dfs(entityset=datetime_es,
target_dataframe_name="transactions",
cutoff_time=pd.Timestamp("2012-3-31 04:00"),
training_window=pd.DateOffset(months=2))
feature_matrix_2, _ = dfs(entityset=datetime_es,
target_dataframe_name="transactions",
cutoff_time=pd.Timestamp("2012-3-31 04:00"),
training_window=pd.offsets.BDay(44))
assert (feature_matrix.index == [2, 3, 4]).all()
assert (feature_matrix.index == feature_matrix_2.index).all()
def test_accepts_pd_dateoffset_training_window(datetime_es):
feature_matrix, features = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-3-31 04:00"),
training_window=pd.DateOffset(months=2))
feature_matrix_2, features_2 = dfs(
entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-3-31 04:00"),
training_window=pd.offsets.BDay(44))
assert (feature_matrix.index == [2, 3, 4]).all()
assert (feature_matrix.index == feature_matrix_2.index).all()
def test_does_not_warn_with_stacking_feature(pd_es):
with pytest.warns(None) as record:
dfs(entityset=pd_es,
target_entity='régions',
agg_primitives=['percent_true'],
trans_primitives=[GreaterThanScalar(5)],
primitive_options={
'greater_than_scalar': {
'include_entities': ['stores']
}
},
features_only=True)
assert not record
def test_does_not_warn_with_stacking_feature(pd_es):
with pytest.warns(None) as record:
dfs(
entityset=pd_es,
target_dataframe_name="régions",
agg_primitives=["percent_true"],
trans_primitives=[GreaterThanScalar(5)],
primitive_options={
"greater_than_scalar": {"include_dataframes": ["stores"]}
},
features_only=True,
)
assert not record
def test_warns_with_unused_where_primitives(es):
warning_text = "Some specified primitives were not used during DFS:\n" + \
" where_primitives: ['count', 'sum']\n" + \
"This may be caused by a using a value of max_depth that is too small, not setting interesting values, " + \
"or it may indicate no compatible variable types for the primitive were found in the data."
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(entityset=es,
target_entity='customers',
agg_primitives=['count'],
where_primitives=['sum', 'count'],
max_depth=1)
assert record[0].message.args[0] == warning_text
def test_warns_with_unused_custom_primitives(pd_es):
def above_ten(column):
return column > 10
AboveTen = make_trans_primitive(function=above_ten,
input_types=[Numeric],
return_type=Numeric)
trans_primitives = [AboveTen]
warning_text = "Some specified primitives were not used during DFS:\n" + \
" trans_primitives: ['above_ten']\n" + \
"This may be caused by a using a value of max_depth that is too small, not setting interesting values, " + \
"or it may indicate no compatible variable types for the primitive were found in the data."
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(entityset=pd_es,
target_entity='sessions',
trans_primitives=trans_primitives,
max_depth=1)
assert record[0].message.args[0] == warning_text
# Should not raise a warning
with pytest.warns(None) as record:
dfs(entityset=pd_es,
target_entity='customers',
trans_primitives=trans_primitives,
max_depth=1)
def max_above_ten(column):
return max(column) > 10
MaxAboveTen = make_agg_primitive(function=max_above_ten,
input_types=[Numeric],
return_type=Numeric)
agg_primitives = [MaxAboveTen]
warning_text = "Some specified primitives were not used during DFS:\n" + \
" agg_primitives: ['max_above_ten']\n" + \
"This may be caused by a using a value of max_depth that is too small, not setting interesting values, " + \
"or it may indicate no compatible variable types for the primitive were found in the data."
with pytest.warns(UnusedPrimitiveWarning) as record:
dfs(entityset=pd_es,
target_entity='stores',
agg_primitives=agg_primitives,
max_depth=1)
assert record[0].message.args[0] == warning_text
# Should not raise a warning
with pytest.warns(None) as record:
dfs(entityset=pd_es,
target_entity='sessions',
agg_primitives=agg_primitives,
max_depth=1)
def test_accepts_cutoff_time_compose(entities, relationships):
def fraud_occured(df):
return df['fraud'].any()
lm = cp.LabelMaker(target_entity='card_id',
time_index='transaction_time',
labeling_function=fraud_occured,
window_size=1)
transactions_df = entities['transactions'][0]
if isinstance(transactions_df, dd.DataFrame):
transactions_df = transactions_df.compute()
labels = lm.search(transactions_df, num_examples_per_instance=-1)
labels['time'] = pd.to_numeric(labels['time'])
labels.rename({'card_id': 'id'}, axis=1, inplace=True)
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="cards",
cutoff_time=labels)
if isinstance(feature_matrix, dd.DataFrame):
feature_matrix = feature_matrix.compute().set_index('id')
assert len(feature_matrix.index) == 6
assert len(feature_matrix.columns) == len(features) + 1
def test_direct_of_multi_output_transform_feat(es):
class TestTime(TransformPrimitive):
name = "test_time"
input_types = [Datetime]
return_type = Numeric
number_output_features = 6
def get_function(self):
def test_f(x):
times = pd.Series(x)
units = ["year", "month", "day", "hour", "minute", "second"]
return [times.apply(lambda x: getattr(x, unit)) for unit in units]
return test_f
join_time_split = Feature(es["customers"]["signup_date"],
primitive=TestTime)
alt_features = [Feature(es["customers"]["signup_date"], primitive=Year),
Feature(es["customers"]["signup_date"], primitive=Month),
Feature(es["customers"]["signup_date"], primitive=Day),
Feature(es["customers"]["signup_date"], primitive=Hour),
Feature(es["customers"]["signup_date"], primitive=Minute),
Feature(es["customers"]["signup_date"], primitive=Second)]
fm, fl = dfs(
entityset=es,
target_entity="sessions",
trans_primitives=[TestTime, Year, Month, Day, Hour, Minute, Second])
# Get column names of for multi feature and normal features
subnames = DirectFeature(join_time_split, es["sessions"]).get_feature_names()
altnames = [DirectFeature(f, es["sessions"]).get_name() for f in alt_features]
# Check values are equal between
for col1, col2 in zip(subnames, altnames):
assert (fm[col1] == fm[col2]).all()
def test_accepts_single_cutoff_time(entities, relationships):
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=20)
assert len(feature_matrix.index) == 5
assert len(feature_matrix.columns) == len(features)
def test_accepts_pandas_training_window(datetime_es):
feature_matrix, features = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-4-1 04:00"),
training_window=pd.Timedelta(90, "D"))
assert (feature_matrix.index == [2, 3, 4]).all()
def test_accepts_no_cutoff_time(entities, relationships):
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
instance_ids=[1, 2, 3, 5, 6])
assert len(feature_matrix.index) == 5
assert len(feature_matrix.columns) == len(features)
def test_passing_strings_to_variable_types_dfs():
variable_types = find_variable_types()
teams = pd.DataFrame({
'id': range(3),
'name': ['Breakers', 'Spirit', 'Thorns']
})
games = pd.DataFrame({
'id': range(5),
'home_team_id': [2, 2, 1, 0, 1],
'away_team_id': [1, 0, 2, 1, 0],
'home_team_score': [3, 0, 1, 0, 4],
'away_team_score': [2, 1, 2, 0, 0]
})
entities = {
'teams': (teams, 'id', None, {
'name': 'natural_language'
}),
'games': (games, 'id')
}
relationships = [('teams', 'id', 'games', 'home_team_id')]
features = dfs(entities,
relationships,
target_entity="teams",
features_only=True)
name_class = features[0].entity['name'].__class__
assert name_class == variable_types['natural_language']
def test_accepts_cutoff_time_compose(dataframes, relationships):
def fraud_occured(df):
return df["fraud"].any()
lm = cp.LabelMaker(
target_dataframe_name="card_id",
time_index="transaction_time",
labeling_function=fraud_occured,
window_size=1,
)
transactions_df = to_pandas(dataframes["transactions"][0])
labels = lm.search(transactions_df, num_examples_per_instance=-1)
labels["time"] = pd.to_numeric(labels["time"])
labels.rename({"card_id": "id"}, axis=1, inplace=True)
feature_matrix, features = dfs(
dataframes=dataframes,
relationships=relationships,
target_dataframe_name="cards",
cutoff_time=labels,
)
feature_matrix = to_pandas(feature_matrix, index="id")
assert len(feature_matrix.index) == 6
assert len(feature_matrix.columns) == len(features) + 1
def test_passing_strings_to_logical_types_dfs():
teams = pd.DataFrame({
"id": range(3),
"name": ["Breakers", "Spirit", "Thorns"]
})
games = pd.DataFrame({
"id": range(5),
"home_team_id": [2, 2, 1, 0, 1],
"away_team_id": [1, 0, 2, 1, 0],
"home_team_score": [3, 0, 1, 0, 4],
"away_team_score": [2, 1, 2, 0, 0],
})
dataframes = {
"teams": (teams, "id", None, {
"name": "natural_language"
}),
"games": (games, "id"),
}
relationships = [("teams", "id", "games", "home_team_id")]
features = dfs(dataframes,
relationships,
target_dataframe_name="teams",
features_only=True)
name_logical_type = features[0].dataframe["name"].ww.logical_type
assert isinstance(name_logical_type, NaturalLanguage)
def test_passing_strings_to_logical_types_dfs():
teams = pd.DataFrame({
'id': range(3),
'name': ['Breakers', 'Spirit', 'Thorns']
})
games = pd.DataFrame({
'id': range(5),
'home_team_id': [2, 2, 1, 0, 1],
'away_team_id': [1, 0, 2, 1, 0],
'home_team_score': [3, 0, 1, 0, 4],
'away_team_score': [2, 1, 2, 0, 0]
})
dataframes = {
'teams': (teams, 'id', None, {
'name': 'natural_language'
}),
'games': (games, 'id')
}
relationships = [('teams', 'id', 'games', 'home_team_id')]
features = dfs(dataframes,
relationships,
target_dataframe_name="teams",
features_only=True)
name_logical_type = features[0].dataframe['name'].ww.logical_type
assert isinstance(name_logical_type, NaturalLanguage)
def test_accepts_single_cutoff_time(entities, relationships):
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=20)
assert len(feature_matrix.index) == 6
assert len(feature_matrix.columns) == len(features)
def test_accepts_no_cutoff_time(entities, relationships):
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
instance_ids=[1, 2, 3, 5, 6])
assert len(feature_matrix.index) == 5
assert len(feature_matrix.columns) == len(features)
def test_direct_features_of_multi_output_agg_primitives(pd_es):
class ThreeMostCommonCat(AggregationPrimitive):
name = "n_most_common_categorical"
input_types = [ColumnSchema(semantic_tags={"category"})]
return_type = ColumnSchema(semantic_tags={"category"})
number_output_features = 3
def get_function(self, agg_type="pandas"):
def pd_top3(x):
counts = x.value_counts()
counts = counts[counts > 0]
array = np.array(counts.index[:3])
if len(array) < 3:
filler = np.full(3 - len(array), np.nan)
array = np.append(array, filler)
return array
return pd_top3
fm, fl = dfs(
entityset=pd_es,
target_dataframe_name="log",
agg_primitives=[ThreeMostCommonCat],
trans_primitives=[],
max_depth=3,
)
has_nmost_as_base = []
for feature in fl:
is_base = False
if len(feature.base_features) > 0 and isinstance(
feature.base_features[0].primitive, ThreeMostCommonCat):
is_base = True
has_nmost_as_base.append(is_base)
assert any(has_nmost_as_base)
true_result_rows = []
session_data = {
0: ["coke zero", "car", np.nan],
1: ["toothpaste", "brown bag", np.nan],
2: ["brown bag", np.nan, np.nan],
3: set(["Haribo sugar-free gummy bears", "coke zero", np.nan]),
4: ["coke zero", np.nan, np.nan],
5: ["taco clock", np.nan, np.nan],
}
for i, count in enumerate([5, 4, 1, 2, 3, 2]):
while count > 0:
true_result_rows.append(session_data[i])
count -= 1
tempname = "sessions.N_MOST_COMMON_CATEGORICAL(log.product_id)[%s]"
for i, row in enumerate(true_result_rows):
for j in range(3):
value = fm[tempname % (j)][i]
if isinstance(row, set):
assert pd.isnull(value) or value in row
else:
assert (pd.isnull(value)
and pd.isnull(row[j])) or value == row[j]
def test_accepts_pd_timedelta_training_window(datetime_es):
# TODO: Update to use Dask entities when issue #882 is closed
feature_matrix, features = dfs(entityset=datetime_es,
target_entity="transactions",
cutoff_time=pd.Timestamp("2012-3-31 04:00"),
training_window=pd.Timedelta(61, "D"))
assert (feature_matrix.index == [2, 3, 4]).all()
def test_accepts_cutoff_time_df(entities, relationships):
cutoff_times_df = pd.DataFrame({"instance_id": [1, 2, 3],
"time": [10, 12, 15]})
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df)
assert len(feature_matrix.index) == 3
assert len(feature_matrix.columns) == len(features)
def test_dask_kwargs(entities, relationships):
cutoff_times_df = pd.DataFrame({"instance_id": [1, 2, 3],
"time": [10, 12, 15]})
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df)
with cluster() as (scheduler, [a, b]):
dask_kwargs = {'cluster': scheduler['address']}
feature_matrix_2, features_2 = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df,
dask_kwargs=dask_kwargs)
assert features == features_2
for column in feature_matrix:
for x, y in zip(feature_matrix[column], feature_matrix_2[column]):
assert ((pd.isnull(x) and pd.isnull(y)) or (x == y))
def test_ignores_instance_ids_if_cutoff_df(entities, relationships):
cutoff_times_df = pd.DataFrame({"instance_id": [1, 2, 3],
"time": [10, 12, 15]})
instance_ids = [1, 2, 3, 4, 5]
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df,
instance_ids=instance_ids)
assert len(feature_matrix.index) == 3
assert len(feature_matrix.columns) == len(features)
def test_all_variables(entities, relationships):
cutoff_times_df = pd.DataFrame({"instance_id": [1, 2, 3],
"time": [10, 12, 15]})
instance_ids = [1, 2, 3, 4, 5]
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df,
instance_ids=instance_ids,
agg_primitives=[Max, Mean, Min, Sum],
trans_primitives=[],
max_depth=3,
allowed_paths=None,
ignore_entities=None,
ignore_variables=None,
seed_features=None)
assert len(feature_matrix.index) == 3
assert len(feature_matrix.columns) == len(features)
def test_approximate_features(entities, relationships):
cutoff_times_df = pd.DataFrame({"instance_id": [1, 3, 1, 5, 3, 6],
"time": [11, 16, 16, 26, 17, 22]})
feature_matrix, features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
cutoff_time=cutoff_times_df,
approximate=5,
cutoff_time_in_index=True)
direct_agg_feat_name = 'cards.PERCENT_TRUE(transactions.fraud)'
assert len(feature_matrix.index) == 6
assert len(feature_matrix.columns) == len(features)
truth_index = pd.MultiIndex.from_arrays([[1, 3, 1, 5, 3, 6],
[11, 16, 16, 26, 17, 22]],
names=('id', 'time'))
truth_values = pd.Series(data=[1.0, 0.5, 0.5, 1.0, 0.5, 1.0],
index=truth_index)
truth_values.sort_index(level='time', kind='mergesort', inplace=True)
assert (feature_matrix[direct_agg_feat_name] == truth_values).all()
def test_mock_customer():
es = load_mock_customer(return_entityset=True)
fm, fl = dfs(entityset=es, target_entity="customers", max_depth=3)
for feature in fl:
assert feature.get_name() in fm.columns
def test_features_only(entities, relationships):
features = dfs(entities=entities,
relationships=relationships,
target_entity="transactions",
features_only=True)
assert len(features) > 0
