Wrong feature_ts when calculating batch features about fraudfinder HOT 4 CLOSED

Closing this issue, but feel free to re-open if it is still unclear -- thank you!

from fraudfinder.

This is what I get in the customers table the feature_ts make none sense to me. Each row should have each own different feature_ts value.

from fraudfinder.

One more thing:
I changed this

PARSE_TIMESTAMP('%Y-%m-%d %H:%M:%S', CONCAT(@END_DATE_TRAIN, ' ', STRING(TIME_TRUNC(CURRENT_TIME(), SECOND))), 'UTC') AS feature_ts,

for this

a.TX_TS AS feature_ts,

and looks like now the features have the correct time

from fraudfinder.

Hi @artefactop,

Thanks for reaching out to the FraudFinder team.

We want you to know that we recognized the issue related with the query variable and the date replacement within the backfill function. Also we verified the feature engineering process and we reproduced the incorrected behavior where each row of the entity tables have each row with the same feature_ts value.

Thanks to your feedback, we decided to simplify the feature engineering process which was originally designed to simulate a scheduled daily process. In the upcoming release, aggregated feature values will be computed based on transaction timestamp. Consequently, you will create new feature values in your feature store using the transaction timestamp. And a new batch of features will be retrieved by point-in-time query which will find the most up-to-date feature values looking up the transaction timestamp and the label timestamp.

Best regards,

FraudFinder Team

from fraudfinder.