Testing On-Demand Features

Import libraries and select your workspace

import tecton
import pandas
from datetime import datetime

ws = tecton.get_workspace("prod")

On-Demand Feature Views with no Feature View Dependencies

Load an On-Demand Feature View

fv = ws.get_feature_view("transaction_amount_is_high")
fv.summary()

Execute an On-Demand Feature View Online

Because this On-Demand Feature View has no Batch or Stream Feature View dependencies it only needs request data to execute.

fv.get_online_features(request_data={"amt": 17000}).to_dict()

Out: {"transaction_amount_is_high": True}

Execute an On-Demand Feature View Offline

Create a spine DataFrame with the request data to transform.

spine_df = pandas.DataFrame({"amt": [200, 20]})
features_df = fv.get_historical_features(spine=spine_df).to_pandas()
display(features_df)

	amt	transaction_amount_is_high__transaction_amount_is_high
0	200	True
1	20	False

On-Demand Feature Views with Feature View Dependencies

Load an On-Demand Feature View

fv = ws.get_feature_view("transaction_amount_is_higher_than_average")

Run Feature View Transformation Pipeline with Mock Inputs

To use run on On-Demand Feature View, all Feature View inputs must be mocked.

transaction_request = pandas.DataFrame([{"amt": 100.0}])

user_transaction_amount_metrics = pandas.DataFrame([{"amt_mean_1d_10m": 200.0}])

result = fv.run(
    transaction_request=transaction_request,
    user_transaction_amount_metrics=user_transaction_amount_metrics,
)
print(result)

{"transaction_amount_is_higher_than_average": 0}

Execute On-Demand FeatureView with Dependencies Offline

Because this On-Demand Feature View depends on another Feature View, we will first preview the dependent Feature View. We can use this to select keys that will be needed for historical lookups in order for the On-Demand Feature View to run. The dependent column in the Feature View below is amt_mean_1d_10m.

dependent_fv = ws.get_feature_view("user_transaction_amount_metrics")
result_dataframe = dependent_fv.get_historical_features(
    start_time=datetime(2022, 5, 1), end_time=datetime(2022, 5, 2), from_source=True
).to_pandas()
display(result_dataframe)

	user_id	timestamp	amt_sum_1h_10m	amt_sum_1d_10m	amt_sum_3d_10m	amt_mean_1h_10m	amt_mean_1d_10m	amt_mean_3d_10m	_effective_timestamp
0	user_222506789984	2022-05-01 21:10:00	120.66	120.66	1404.69	120.66	120.66	351.173	2022-05-01 21:10:00
1	user_26990816968	2022-05-01 19:50:00	134.73	134.73	144.07	134.73	134.73	72.035	2022-05-01 19:50:00
2	user_337750317412	2022-05-01 02:00:00	76.45	76.45	181.4	76.45	76.45	60.4667	2022-05-01 02:00:00
3	user_337750317412	2022-05-01 07:20:00	2.13	78.58	183.53	2.13	39.29	45.8825	2022-05-01 07:20:00
4	user_337750317412	2022-05-01 14:10:00	9.96	88.54	193.49	9.96	29.5133	38.698	2022-05-01 14:10:00

Create a spine DataFrame with events to look up. For more information on spines, check out Selecting Sample Keys and Timestamps.

spine_df = pandas.DataFrame(
    {
        "user_id": [
            "user_26990816968",
            "user_337750317412",
            "user_222506789984",
            "user_337750317412",
        ],
        "timestamp": [
            datetime(2022, 5, 1, 19, 51),
            datetime(2022, 5, 2, 5, 0),
            datetime(2022, 5, 1, 21, 11),
            datetime(2022, 5, 1, 7, 21),
        ],
        "amt": [71.82, 80.98, 180.55, 66.57],
    }
)
display(spine_df)

	user_id	timestamp	amt
0	user_26990816968	2022-05-01 19:51:00	71.82
1	user_337750317412	2022-05-02 05:00:00	80.98
2	user_222506789984	2022-05-01 21:11:00	180.55
3	user_337750317412	2022-05-01 07:21:00	66.57

result_dataframe = fv.get_historical_features(spine=spine_df, from_source=True).to_pandas()
display(result_dataframe)

	user_id	timestamp	amt	transaction_amount_is_higher_than_average__transaction_amount_is_higher_than_average
0	user_26990816968	2022-05-01 19:51:00	71.82	0
1	user_337750317412	2022-05-02 05:00:00	80.98	1
2	user_222506789984	2022-05-01 21:11:00	180.55	1
3	user_337750317412	2022-05-01 07:21:00	66.57	1

Execute On-Demand FeatureView with Dependencies Online

# This will compare a transaction amount against the user's most recent average 24h transaction amount.
fv.get_online_features(request_data={"amt": 150}, join_keys={"user_id": "user_724235628997"}).to_dict()

Out: {"transaction_amount_is_higher_than_average": 1}