Simulated Delay Latency for Offline Online Feature Parity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional techniques for offline near real-time (NRT) feature simulation fail to accurately reflect the complexity of event time delays and concurrent data processing in online environments, leading to model refits and prolonged time-to-market for machine-learning models.

Innovation Solution

A computing device simulates delay latency associated with event processing in an online environment and provides this latency to a machine-learning model, enabling improved offline/online feature match rates by generating enriched event data with time stamps, which is used to train the model before deployment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional offline NRT feature simulation techniques are used, then the machine-learning model can be trained with offline data, but the model fails to accurately reflect online event time delays and concurrent data processing complexity

Engineering Contradiction:
Improvemodel accuracyVSAvoidevent time delay information
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The system performs preliminary action by simulating online event processing delays and concurrent data processing conditions during the offline training phase. Time stamps are assigned to events and processed through a simulation engine that introduces artificial delays matching online processing characteristics, allowing the model to learn temporal patterns before deployment without requiring actual online data

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates a copy of the online processing environment within the offline simulation framework. By replicating event processing pipelines, time delay characteristics, and data flow patterns from the online system, the offline model receives training data that mirrors real-world conditions, improving its ability to generalize to production environments

Inventive Principle:
Principle #26Copying

2Productivity

If conventional offline simulation is used, then training data can be generated efficiently, but the model requires refits when deployed online due to timing mismatches

Engineering Contradiction:
Improvetraining efficiencyVSAvoidonline/offline feature match rate
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The system applies parameter changes by dynamically adjusting time delay parameters and processing speed parameters in the offline simulation to match online characteristics. The simulation engine modifies event processing timing, batch size parameters, and aggregation window parameters to reflect real-world online processing conditions, enabling the model to adapt to production environments without refitting

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If offline NRT feature simulation is performed without simulating processing delays, then the simulation process is simple and fast, but the model does not generalize well to online environments

Engineering Contradiction:
Improvesimulation simplicityVSAvoidfeature match precision
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The system introduces an intermediary simulation engine that bridges the gap between simple offline simulation and complex online processing. This intermediary layer processes events through virtual pipelines that introduce controlled delays and transformations, maintaining simulation simplicity while improving feature match precision through realistic temporal modeling

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4446950A1Near real-time feature simulation for online/offline point-in-time data parity
Publication Date: 2024.10.16 EBAY INC
  • EP4446950A1 patent drawingFigure 1
  • EP4446950A1 patent drawingFigure 2
  • EP4446950A1 patent drawingFigure 3

AI summary

Near real-time feature simulation for online/offline point-in-time data parity is described. A computing device may assign, to respective events from a series of events, a series of time stamps associated with a near real-time (NRT) variable. The computing device may simulate a delay latency associated with processing the respective events via an online processing environment based on the series of time stamps. The computing device may provide the series of events and the simulated delay latency to a machine-learning model configured to model an outcome of the series of events using the simulated delay latency.