Machine learning platform and pipeline for efficient data processing
A system enables agile model development to speed up innovation by data scientists. Model training and deployment are coordinated and standardized to reduce redundancy. Data is obtained for feature generation and reformatted and de-sensitized for storage. The features are stored in locations available to all models and training modules of a system so data does not need to be adjusted for new models. To generate a machine learning model, the system establishes a cohort for evaluation by the model. A model template and features for use by the model are identified. The selected template and features are used for experimentation and evaluation. Model training artifacts, such as model weights are subsequently recorded in a model store and the model scripts and settings can then be registered in a centralized database where it can be accessed for execution.
1 . A computer-implemented method for generating a standardized predictive model, the method comprising:
selecting a set of input features from a generated and standardized feature set, the generated and standardized feature set being stored in a centralized feature store accessible to a plurality of models and comprising (i) a training feature set including normalized features from before a score date and (ii) a prediction feature set including normalized features from on or after the score date;
designating input and output formats for the predictive model;
generating a set of templates for training, scoring, and monitoring the predictive model using a template generator to output model templates and feature templates having executable placeholders for specifying the selected input features and the designated output formats;
performing one or more model training experiments to generate a set of trained models, including training using the training feature set and evaluating using the prediction feature set;
identifying a preferred predictive model from the set of trained models to be used in production;
registering the identified preferred predictive model in a central registry that (i) is accessible to a plurality of scripts, (ii) records trained model artifacts, including model weights, and (iii) separately registers a configuration file comprising configuration parameters for the identified preferred predictive model; and
automating a regular scoring and monitoring of the identified preferred predictive model by scheduling production executions on future data sets and, for each execution, performing (i) an upstream dependency check of required inputs and (ii) a score validation of outputs with centralized success/failure logging.
2 . The computer-implemented method of claim 1 , wherein generating the set of templates for training, scoring, and monitoring the predictive model comprises editing stored template documentation to expose relevant functions.
3 . The computer-implemented method of claim 1 , wherein a template includes executable code with placeholders that is edited by a developer.
4 . The computer-implemented method of claim 1 , wherein the selected set of input features are selected from a feature store that includes user feature data.
5 . The computer-implemented method of claim 4 , wherein the feature store includes data sets for features that are updated daily and data sets for features that are updated monthly.
6 . The computer-implemented method of claim 1 , wherein registering the identified preferred predictive model in the central registry comprises storing the identified preferred predictive model in a data storage location that is accessible to the plurality of scripts.
7 . The computer-implemented method of claim 1 , wherein automating the regular scoring and monitoring of the identified preferred predictive model comprises scheduling deployment of the identified preferred predictive model to be applied to future data sets.
8 . The computer-implemented method of claim 1 , wherein automating the regular scoring and monitoring of the identified preferred predictive model comprises deploying the identified preferred predictive model to a centralized production workspace.
9 . The computer-implemented method of claim 1 , further comprising registering the configuration file including the configuration parameters for the identified preferred predictive model in a separate centralized location.
10 . The computer-implemented method of claim 1 , wherein the scripts used to maintain and deploy the identified preferred predictive model are stored independently from the identified preferred predictive model, and wherein the scripts used to maintain the identified preferred predictive model are accessible for use by a plurality of trained machine learning models.