
Ultimate access to all questions.
In the context of a ticketing platform for a renowned cinema chain, customers utilize a mobile app to search and purchase tickets. The platform employs a Dataflow streaming pipeline to handle requests, which includes verifying ticket availability, assigning prices, processing payments, reserving tickets, and updating the database upon successful purchases. Each step must maintain a low latency of no more than 50 milliseconds. To enhance the likelihood of ticket purchases, a logistic regression model has been developed using BigQuery ML to offer promotional codes for free popcorn. Given the constraints of low latency and the need for real-time processing, which of the following methods is the simplest and most effective to deploy this model into production? Choose the best option.
A
Execute batch inference with BigQuery ML every five minutes on each new set of tickets issued, despite the potential delay in offering promotional codes.
B
Export your model in TensorFlow format, and add a tfx_bsl.public.beam.RunInference step to the Dataflow pipeline, which may introduce additional processing steps and latency._
C
Transfer your model in TensorFlow format, deploy it on Vertex AI, and query the prediction endpoint from your streaming pipeline, considering the network latency and the cost of Vertex AI endpoints.
D
Transform your model with TensorFlow Lite (TFLite), and add it to the mobile app so that the promo code and the incoming request arrive together in Pub/Sub, ensuring minimal latency and real-time processing.
E
Both B and C are viable options but consider the trade-offs between additional processing steps and network latency.