Platform-Agnostic Predictive Models via SQL Interchange
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive analytics solutions lack a standardized, flexible way to export predictive models from training environments to production environments, leading to slow turnaround and errors due to platform compatibility issues and the limitations of existing interchange formats like PMML and PFA, which are not adequately flexible to encapsulate complex data transformations.
Innovation Solution
Utilizing SQL as the basis for creating a predictive model interchange format, which is already widely supported and flexible enough to encapsulate sequences of data transformations, enabling the generation of database management system instructions that can be executed across different computing platforms, thereby enabling platform-agnostic predictive models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If PMML is used as the interchange format for predictive models, then the predictive model can be exported between different computing platforms, but the format lacks the flexibility to specify complex pre-processing and post-processing data transformations
Solution Approach 1:
The patent uses SQL as the interchange format, which is a universal language already widely supported across different computing platforms and database systems. SQL can express both the predictive model logic and the complex data transformations (pre-processing and post-processing) in a single standardized format, eliminating the need for separate transformation specifications and manual augmentation.
Solution Approach 2:
The patent introduces an interchange tool that acts as a mediator to automatically translate predictive model definitions from one platform's format into SQL instructions. This intermediary tool handles the complexity of format conversion and transformation specification, allowing seamless export between different computing platforms while maintaining full functionality.
2Productivity
If manual translation of predictive analytics results is performed from training system to production system, then the predictive model can be deployed, but the process is slow and error-prone
Solution Approach 1:
The system enables self-service automated export where the interchange tool automatically translates predictive model definitions into platform-agnostic SQL instructions without requiring manual intervention. The tool handles the entire translation process, generating executable SQL code that can be directly deployed to production systems, significantly reducing both time and errors compared to manual translation methods.
3Adaptability or versatility
If different computing platforms are used for training and production environments, then organizational flexibility is improved, but platform compatibility issues arise when exporting predictive models
Solution Approach 1:
The patent establishes SQL as a universal intermediary language that works across all computing platforms. Since SQL is already natively supported by virtually all database systems and computing platforms, the predictive model can be exported from any training platform to any production platform without compatibility issues. The interchange tool translates the model definition into standard SQL instructions that execute uniformly across different platforms.
Data Source
AI summary
Platform-agnostic predictive models based on database management system instructions are described. A system identifies a representation of data transformations associated with a first predictive model that executes on a first computing platform. The system parses the representation of data transformations. The system generates database management system instructions that correspond to the parsed representation of data transformations. The system sends the database management system instructions to a second predictive model that executes on a second computing platform, thereby enabling the second predictive model to execute at least some of the database management system instructions to generate a prediction. The first computing platform and the second computing platform are different types of computing platforms.