def test_search_with_undefined_labels(transactions, total_spent_fn):
def total_spent(ds):
return total_spent_fn(ds) % 3
lm = LabelMaker(target_entity='customer_id',
time_index='time',
labeling_function=total_spent)
n_examples = {1: 1, 2: 1}
given_labels = lm.search(transactions,
num_examples_per_instance=n_examples,
gap=1)
given_labels = to_csv(given_labels, index=False)
labels = [
'customer_id,time,total_spent',
'0,2019-01-01 08:00:00,2',
'0,2019-01-01 08:30:00,1',
'1,2019-01-01 09:30:00,2',
'1,2019-01-01 10:00:00,1',
'2,2019-01-01 10:30:00,1',
'2,2019-01-01 11:30:00,2',
'3,2019-01-01 12:30:00,1',
]
assert given_labels == labels
def test_label_type(transactions, total_spent_fn):
lm = LabelMaker(target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn)
lt = lm.search(transactions, num_examples_per_instance=1)
assert lt.target_types['total_spent'] == 'continuous'
assert lt.bin(2).target_types['total_spent'] == 'discrete'
def test_search_offset_mix_2(transactions, total_spent_fn):
"""
Test offset mix with window_size (absolute), minimum_data (relative), and gap (absolute).
"""
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size='30min',
)
given_labels = lm.search(
transactions,
num_examples_per_instance=2,
minimum_data=2,
verbose=False,
)
given_labels = to_csv(given_labels, index=False)
labels = [
'customer_id,cutoff_time,total_spent',
'1,2019-01-01 10:00:00,1',
'2,2019-01-01 11:30:00,1',
'2,2019-01-01 12:00:00,1',
]
assert given_labels == labels
def test_search_offset_mix_7(transactions, total_spent_fn):
"""
Test offset mix with window_size (relative), minimum_data (relative), and gap (relative).
"""
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size=10,
)
given_labels = lm.search(
transactions,
num_examples_per_instance=float('inf'),
verbose=False,
)
given_labels = to_csv(given_labels, index=False)
labels = [
'customer_id,cutoff_time,total_spent',
'0,2019-01-01 08:00:00,2',
'1,2019-01-01 09:00:00,3',
'2,2019-01-01 10:30:00,4',
'3,2019-01-01 12:30:00,1',
]
assert given_labels == labels
def test_search_offset_mix_3(transactions, total_spent_fn):
"""
Test offset mix with window_size (absolute), minimum_data (absolute), and gap (relative).
"""
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size='8h',
)
given_labels = lm.search(
transactions,
num_examples_per_instance=-1,
minimum_data='2019-01-01 08:00:00',
gap=1,
)
given_labels = to_csv(given_labels, index=False)
labels = [
'customer_id,time,total_spent',
'0,2019-01-01 08:00:00,2',
'0,2019-01-01 08:30:00,1',
'1,2019-01-01 09:00:00,3',
'1,2019-01-01 09:30:00,2',
'1,2019-01-01 10:00:00,1',
'2,2019-01-01 10:30:00,4',
'2,2019-01-01 11:00:00,3',
'2,2019-01-01 11:30:00,2',
'2,2019-01-01 12:00:00,1',
'3,2019-01-01 12:30:00,1',
]
assert given_labels == labels
def test_column_based_windows(transactions, total_spent_fn):
session_id = [1, 2, 3, 3, 4, 5, 5, 5, 6, 7]
df = transactions.assign(session_id=session_id)
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
window_size='session_id',
labeling_function=total_spent_fn,
)
actual = lm.search(df, -1).pipe(to_csv, index=False)
expected = [
'customer_id,time,total_spent',
'0,2019-01-01 08:00:00,1',
'0,2019-01-01 08:30:00,1',
'1,2019-01-01 09:00:00,2',
'1,2019-01-01 10:00:00,1',
'2,2019-01-01 10:30:00,3',
'2,2019-01-01 12:00:00,1',
'3,2019-01-01 12:30:00,1',
]
assert actual == expected
def test_label_type(transactions, total_spent_fn):
lm = LabelMaker(target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn)
lt = lm.search(transactions,
num_examples_per_instance=1,
label_type='discrete',
verbose=False)
assert lt.label_type == 'discrete'
def test_data_slice_overlap(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size='1h',
)
for ds in lm.slice(transactions, num_examples_per_instance=2):
overlap = ds.index == ds.context.slice_stop
assert not overlap.any()
def test_slice_overlap(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size='1h',
)
for df in lm.slice(transactions, num_examples_per_instance=2):
start, end = df.context.window
is_overlap = df.index == end
assert not is_overlap.any()
def test_search_invalid_n_examples(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
)
with pytest.raises(AssertionError, match='must specify gap'):
next(lm.slice(transactions, num_examples_per_instance=2))
with pytest.raises(AssertionError, match='must specify gap'):
lm.search(transactions, num_examples_per_instance=2)
def test_slice_context(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=total_spent_fn,
window_size='1h',
)
for ds in lm.slice(transactions,
num_examples_per_instance=-1,
verbose=True):
assert ds.context.target_entity == 'customer_id'
assert isinstance(ds.context.target_instance, int)
assert isinstance(ds.context.slice_number, int)
assert ds.context.window == ds.context.gap
def test_search_default(transactions, total_spent_fn):
lm = LabelMaker(target_entity='customer_id', time_index='time', labeling_function=total_spent_fn)
given_labels = lm.search(transactions, num_examples_per_instance=1)
given_labels = to_csv(given_labels, index=False)
labels = [
'customer_id,time,total_spent',
'0,2019-01-01 08:00:00,2',
'1,2019-01-01 09:00:00,3',
'2,2019-01-01 10:30:00,4',
'3,2019-01-01 12:30:00,1',
]
assert given_labels == labels
def test_invalid_threshold(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda: None,
window_size=2,
)
match = 'invalid threshold'
with pytest.raises(ValueError, match=match):
lm.search(
transactions,
num_examples_per_instance=2,
minimum_data=' ',
)
def test_search_on_empty_labels(transactions):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda ds: None,
window_size=2,
)
given_labels = lm.search(
transactions,
minimum_data=1,
num_examples_per_instance=2,
gap=1,
)
assert given_labels.empty
def test_search_offset_negative_1(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda: None,
window_size=2,
)
match = 'offset must be positive'
with pytest.raises(AssertionError, match=match):
lm.search(
transactions,
num_examples_per_instance=2,
minimum_data='-1h',
gap='-1h',
)
def test_search_offset_negative_0(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda: None,
window_size=2,
)
match = 'must be greater than zero'
with pytest.raises(AssertionError, match=match):
lm.search(
transactions,
num_examples_per_instance=2,
minimum_data=-1,
gap=-1,
verbose=False,
)
def test_search_on_empty_data_slices(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda df: None,
window_size=2,
)
transactions = transactions.assign(time=pd.NaT)
given_labels = lm.search(
transactions,
minimum_data=1,
num_examples_per_instance=2,
gap=3,
)
assert given_labels.empty
def test_invalid_offset(transactions, total_spent_fn):
match = 'invalid offset'
with pytest.raises(AssertionError, match=match):
LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda: None,
window_size={},
)
def test_invalid_offset_alias(transactions, total_spent_fn):
match = 'offset must be a valid string'
with pytest.raises(AssertionError, match=match):
LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda: None,
window_size='not an offset alias',
)
def test_search_with_multiple_targets(transactions, total_spent_fn,
unique_amounts_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
window_size=2,
labeling_function={
'total_spent': total_spent_fn,
'unique_amounts': unique_amounts_fn,
},
)
expected = [
'customer_id,time,total_spent,unique_amounts',
'0,2019-01-01 08:00:00,2,1',
'1,2019-01-01 09:00:00,2,1',
'1,2019-01-01 10:00:00,1,1',
'2,2019-01-01 10:30:00,2,1',
'2,2019-01-01 11:30:00,2,1',
'3,2019-01-01 12:30:00,1,1',
]
lt = lm.search(transactions, num_examples_per_instance=-1)
actual = lt.pipe(to_csv, index=False)
info = 'unexpected calculated values'
assert actual == expected, info
expected = [
'customer_id,time,unique_amounts',
'0,2019-01-01 08:00:00,1',
'1,2019-01-01 09:00:00,1',
'1,2019-01-01 10:00:00,1',
'2,2019-01-01 10:30:00,1',
'2,2019-01-01 11:30:00,1',
'3,2019-01-01 12:30:00,1',
]
actual = lt.select('unique_amounts')
actual = actual.pipe(to_csv, index=False)
info = 'selected values differ from calculated values'
assert actual == expected, info
def labels():
df = ft.demo.load_mock_customer(return_single_table=True, random_seed=0)
df = df[['transaction_time', 'customer_id', 'amount']]
df.sort_values('transaction_time', inplace=True)
lm = LabelMaker(
target_entity='customer_id',
time_index='transaction_time',
labeling_function=total_spent,
window_size='1h',
)
lt = lm.search(
df,
minimum_data='10min',
num_examples_per_instance=2,
gap='30min',
drop_empty=True,
verbose=False,
)
lt = lt.threshold(1250)
return lt
def test_search_with_invalid_index(transactions, total_spent_fn):
lm = LabelMaker(
target_entity='customer_id',
time_index='time',
labeling_function=lambda df: None,
window_size=2,
)
df = transactions.sample(n=10, random_state=0)
match = "data frame must be sorted chronologically"
with pytest.raises(AssertionError, match=match):
lm.search(df, num_examples_per_instance=2)
df = transactions.assign(time=pd.NaT)
match = "index contains null values"
with pytest.raises(AssertionError, match=match):
lm.search(df, num_examples_per_instance=2)