Predictive Modeling Server System for Efficient Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive models require large volumes of training data and expertise to select and tune effectively, which can be resource-intensive for client computing systems, and there is a need for a method to efficiently train and select predictive models that can handle diverse input data types.
Innovation Solution
A predictive modeling server system that receives training data from client computing systems, trains multiple models using various functions and hyper-parameter configurations, generates scores for model effectiveness, and selects the best model for client use, allowing remote training and prediction operations without expending client-side resources, and enabling integration of new knowledge without additional support.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple predictive models are trained with different training functions and hyper-parameter configurations to improve prediction accuracy, then the model selection quality improves, but the computational resources and time required for training increase significantly
Solution Approach 1:
The system pre-trains multiple predictive models with different training functions and hyper-parameter configurations and stores them in a model repository before they are needed. When a prediction request arrives, the system retrieves pre-trained models rather than training them on-demand, significantly reducing the time loss while maintaining the ability to provide accurate predictions through model selection.
Solution Approach 2:
The model training process is segmented into multiple independent model training operations, where different models are trained separately with different configurations. This allows parallel training of multiple models, distributing the computational load and reducing the overall time required compared to sequential training of a single model with multiple configurations.
2Measurement precision
If multiple predictive models are trained and evaluated to select the best model, then the prediction quality improves, but the computational resources required increase
Solution Approach 1:
Instead of training and evaluating all possible model configurations, the system trains a representative subset of models with different training functions and hyper-parameter configurations. This partial action approach provides sufficient model selection quality without the excessive computational resource consumption that would result from exhaustively training all possible configurations.
Solution Approach 2:
Models are pre-trained and stored in a repository before prediction requests arrive. This preliminary action allows the system to have multiple trained models ready for selection, reducing the computational resources needed at prediction time while maintaining the ability to select high-quality models for accurate predictions.
3Adaptability or versatility
If a client computing system trains predictive models locally with large volumes of training data, then the model customization improves, but the resource burden on the client system increases
Solution Approach 1:
A server system acts as an intermediary between the client computing system and the model training process. The client provides training data and receives trained models through this intermediary, which handles the computationally intensive training operations. This allows the client to obtain customized models without bearing the resource burden of local training.
Solution Approach 2:
The model training function is extracted from the client computing system and relocated to a server system. This extraction removes the resource burden of training from the client while preserving the ability to obtain customized models trained on the client's data through the server's training capabilities.
4Loss of energy
If existing predictive models are used without retraining, then the resource consumption is low, but the models cannot adapt to new knowledge or data patterns
Solution Approach 1:
The system pre-trains multiple models with different configurations and stores them in a repository. When new knowledge or data patterns need to be incorporated, new models can be pre-trained with updated training data and added to the repository without requiring retraining of existing models or interruption of service. This maintains low resource consumption during operation while enabling adaptability through model updates.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on one or more computer storage devices, for training a predictive model. In one aspect, a method includes receiving over a network predictive modeling training data from a client computing system. The training data and multiple training functions obtained from a repository of training functions are used to train multiple predictive models. A score is generated for each of the trained predictive models, where each score represents an estimation of the effectiveness of the respective trained predictive model. A first trained predictive model is selected from among the trained predictive models based on the generated scores. Access to the first trained predictive model is provided to the client computing system.