dressipi_nvt.py

from merlin.datasets.synthetic import generate_data
train, valid = generate_data("dressipi2022-preprocessed", num_rows=10000, set_sizes=(0.8, 0.2))
item_features_names = ['f_' + str(col) for col in [47, 68]]
cat_features = [['item_id', 'purchase_id']] + item_features_names >> nvt.ops.Categorify()

features = ['session_id', 'timestamp', 'date'] + cat_features
to_aggregate = {}
to_aggregate['date'] = ["first"]
to_aggregate['item_id'] = ["last", "list"]
to_aggregate['purchase_id'] = ["first"]   

for name in item_features_names: 
    to_aggregate[name] = ['list']

groupby_features = features >> nvt.ops.Groupby(
groupby_cols=["session_id"], 
sort_cols=["date"],
aggs= to_aggregate,
name_sep="_")

item_last = (
    groupby_features['item_id_last'] >> 
    AddMetadata(tags=[Tags.ITEM, Tags.ITEM_ID])
)
item_list = (
    groupby_features['item_id_list'] >> 
    AddMetadata(
        tags=[Tags.ITEM, Tags.ITEM_ID, Tags.LIST, Tags.SEQUENCE]
    )
)
feature_list = (
    groupby_features[[name+'_list' for name in item_features_names]] >> 
    AddMetadata(
        tags=[Tags.SEQUENCE, Tags.ITEM, Tags.LIST]
    )
)

other_features = groupby_features['session_id', 'date_first']

groupby_features = item_last + item_list + feature_list + other_features + groupby_features['purchase_id_first']


list_features = [name+'_list' for name in item_features_names] + ['item_id_list']
nonlist_features = ['session_id', 'date_first', 'item_id_last', 'purchase_id_first']


SESSIONS_MAX_LENGTH = 3
truncated_features = groupby_features[list_features] >> nvt.ops.ListSlice(-SESSIONS_MAX_LENGTH) >> nvt.ops.Rename(postfix = '_seq')

final_features = groupby_features[nonlist_features] + truncated_features


workflow = nvt.Workflow(final_features)
# fit data
train = workflow.fit_transform(train)