Ufdjrw

I am using feature tools to create monthly aggregations.

I have toy data consisting of loan applications (1000 ID_APPLICATION; 1000 TIME_APPLICATION)
and 200k transactions (-> ~200 transactions for 1 person; 1 transaction have information like AMOUNT, TIME and other, not needed for this example). TIME column consists of ~200 different times for one person, in previous year or more.

constants.py
____________
ID_APPLICATION_COLUMN = "ID_APPLICATION"
ID_TRANSACTIONS_COLUMN = "ID_TRANSACTION"
TIME_COLUMN = "TIME"
TIME_APPLICATION_COLUMN = "TIME_APPLICATION"
ENTITY_SET_NAME = "clients"
TRANSACTIONS_ENTITY_NAME = "transactions"
APPLICATIONS_ENTITY_NAME = "applications"

creation
____________
# we fill the entity_set with the dataframes, and say, which IDs are relevant for given DF
entity_set.entity_from_dataframe(entity_id=cnst.TRANSACTIONS_ENTITY_NAME,
 dataframe=transactions,
 index=cnst.ID_TRANSACTIONS_COLUMN,
 time_index=cnst.TIME_COLUMN)
entity_set.entity_from_dataframe(entity_id=cnst.APPLICATIONS_ENTITY_NAME,
 dataframe=applications,
 index=cnst.ID_APPLICATION_COLUMN,
 time_index=cnst.TIME_APPLICATION_COLUMN)

# Specification of the relationship between entities
r_transactions_applications = ft.Relationship(
 parent_variable=entity_set[cnst.APPLICATIONS_ENTITY_NAME][cnst.ID_APPLICATION_COLUMN],
 child_variable=entity_set[cnst.TRANSACTIONS_ENTITY_NAME][cnst.ID_APPLICATION_COLUMN])
entity_set.add_relationship(r_transactions_applications)

However, I have problem with temporal cutoffs.

when I create them and apply them:

default_agg_primitives = ["count", "sum", "std", "max", "mode", "mean"]
default_trans_primitives = ['month', 'day', 'time_since_previous']
temporal_cutoffs = ft.make_temporal_cutoffs(
 instance_ids=applications[cnst.ID_APPLICATION_COLUMN],
 cutoffs=applications[cnst.TIME_APPLICATION_COLUMN],
 window_size='1m',
 num_windows=6)
transformed_data = ft.dfs(entityset=entity_set,
 target_entity=cnst.APPLICATIONS_ENTITY_NAME,
 cutoff_time=temporal_cutoffs,
 cutoff_time_in_index=True,
 trans_primitives=default_trans_primitives,
 agg_primitives=default_agg_primitives,
 max_depth=2)

As I am aggregating for the application level, I get 1000 rows without temporal cutoffs. What I get when I apply them is 6000 rows, however 5000 rows (all months before, except for the last one) are 0 or NaN, and the rest is just the same as if I would not be using temporal cutoffs at all.

To me, it seems that the TIME column is not registered and the dataset is not splitted.

Where can I set this?