Joint ML Pipeline Optimization via Shared Data Transformations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning pipeline optimization is hindered by redundant and unnecessary computations, especially when dealing with large datasets and complex multivariate time series data, where model performance often saturates with increased data and random partitioning or ordering is not applicable due to varying asset characteristics and sequential nature of time series data.
Innovation Solution
A method for joint optimization of multi-data machine learning pipelines that allocates training data incrementally using asset and failure-centric information, identifies common data transformations across pipelines, and executes them only once to reduce duplicative computations, thereby improving performance and reducing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple ML pipelines are executed independently to evaluate different models, then model performance can be accurately assessed, but computational costs increase and processing time is extended due to redundant data transformations
Solution Approach 1:
The patent merges multiple ML pipeline executions by identifying and factoring out common data transformation operations. When multiple pipelines share common transformations, the system executes these transformations once and caches the results, then reuses them across multiple pipeline evaluations. This combining approach maintains accurate model performance assessment while eliminating redundant computational operations, directly resolving the contradiction between measurement precision and energy loss.
Solution Approach 2:
The system performs preliminary action by pre-executing common data transformations and caching their results before they are needed by multiple pipelines. The transformation cache stores intermediate results that can be reused across different pipeline evaluations, avoiding repeated computation. This preliminary execution of shared operations reduces computational cost while preserving the ability to accurately assess different model performances.
2Measurement precision
If more training data is used to improve model accuracy, then predictive model accuracy improves, but processing time increases and model performance eventually saturates
Solution Approach 1:
The patent applies partial action by using incremental data sampling strategies. Instead of processing the entire training dataset for every pipeline evaluation, the system uses representative subsets of data that provide sufficient performance metrics. This approach achieves adequate model accuracy assessment without the excessive processing time that would result from using complete datasets, addressing the contradiction between measurement precision and time loss.
Solution Approach 2:
The system performs preliminary data processing and transformation, caching intermediate results that can be reused across multiple pipeline evaluations. By pre-processing common transformations and storing results, the system reduces the actual processing time needed during model evaluation while still using comprehensive training data to ensure accurate performance assessment.
3Measurement precision
If complex data transformations are applied to handle multivariate time series data, then model performance improves, but computational complexity increases
Solution Approach 1:
The patent merges complex data transformation operations that are common across multiple ML pipelines into a single shared execution. By identifying transformations that appear in multiple pipelines and executing them once with results cached for reuse, the system maintains the necessary computational complexity for handling multivariate time series data while reducing the overall system complexity through elimination of redundant operations.
Solution Approach 2:
The system creates universal transformation components that serve multiple pipelines simultaneously. The transformation cache and factorization mechanism provide multi-functional capabilities, where a single transformed dataset can be used as input for multiple different ML pipelines. This universality reduces computational complexity while preserving model performance by allowing complex transformations to be performed once and reused across different modeling approaches.
4Measurement precision
If multiple pipelines are evaluated to find the optimal model, then model selection accuracy improves, but resource usage increases due to duplicative computations
Solution Approach 1:
The patent merges the evaluation of multiple ML pipelines by combining shared data transformation operations into a single execution stream. The transformation cache allows results from one pipeline evaluation to be reused in subsequent evaluations, maintaining accurate model selection capability while significantly reducing resource usage by eliminating duplicative computational operations across the multiple pipeline assessments.
Solution Approach 2:
The system recovers computational resources by caching and reusing transformation results instead of discarding them after each pipeline evaluation. The transformation cache stores intermediate results that can be recovered and reused across multiple pipeline evaluations, reducing the need to re-compute expensive transformations and thereby reducing overall resource usage while maintaining accurate model selection.
Data Source
AI summary
Systems and methods for optimizing and training machine learning (ML) models are provided. In embodiments, a computer implemented method includes: performing, by a processor set, a group execution of ML pipelines using a first subset of a training data set as input data for the ML pipelines, thereby generating a trained ML model for each of the ML pipelines, wherein data transformations that are common between the ML pipelines are implemented only once to generate an output, and the output is shared between the ML pipelines during the group execution of the ML pipelines; generating, by the processor set, performance metrics for each of the trained ML models based on validation data; ranking, by the processor set, the trained ML models based on the performance metrics, thereby generating a list of ranked ML models; and outputting, by the processor set, the list of ranked ML models to a user.