Method and system for developing a machine learning model
A framework for easy development of a machine learning solution is provided. The framework includes connecting data sources at all scale levels from a user; assessing data schema, data risk, and data quality status; performing scalable feature engineering and transformation; and automating machine learning model optimization. The framework also includes functionalities for generating production code and automatic model documentation.
1 . A method for developing at least one model, the method being implemented by at least one processor, the method comprising:
receiving, by the at least one processor via a graphical user interface, at least one input that relates to workflow that includes at least one from among a data engineering workflow and a feature development workflow,
wherein the data engineering workflow includes a data processing task that identifies an answer to at least one question based on a corresponding user characteristic; and
wherein the feature development workflow includes an iterative software development process that outputs an application based on a predeveloped platform;
determining, by the at least one processor via a data assembler, whether data included in the at least one input is in compliance with a predetermined data schema;
generating, by the at least one processor via a data explorer, at least one data diagnostic report that relates to an analysis of at least one characteristic for the at least one input;
generating, by the at least one processor via the data assembler, at least one modeling strategy that relates to a development of at least one feature for the at least one input;
generating, by the at least one processor via the data assembler, the at least one model for the at least one input by using at least one design matrix which comprises a sequential mapping of operations involved in producing a processed dataset and by using multi-stage processor parallelization to determine at least one solution for optimizing the at least one model,
wherein the multi-stage processor parallelization includes separating computation parallelization into multiple processing stages, and
wherein the at least one design matrix relates to at least one explanatory variable from the at least one modeling strategy;
determining, by the at least one processor, a file size of the workflow;
dynamically adjusting, by the at least one processor, a processing paradigm based on the file size,
wherein the processing paradigm relates to a quantity of processing components that are allocated to the workflow,
wherein each of the at least one design matrix includes a collection of at least one feature vector for a plurality of data points, and
wherein each of the plurality of data points correspond to at least one scalar response;
operating the data assembler based on a chain-style syntax that includes arranging operations of the data assembler in a chain; and
utilizing, as input to a subsequent operation of the data assembler, output from a current operation of the data assembler.
2 . The method of claim 1 , further comprising:
automatically optimizing, by the at least one processor via a data model optimizer, the at least one model by generating at least one final model based on the at least one solution.
3 . The method of claim 2 , wherein the optimizing includes using at least one global optimization routine, the at least one global optimization routine including at least one from among a Bayesian optimization routine, a grid search optimization routine, a reinforcement learning optimization routine, a gradient descent optimization routine, a genetic algorithm optimization routine, a simulated annealing optimization routine, a covariance matrix adaptation evolution strategy (CMA-ES) optimization routine, and a Gaussian process optimization routine.
4 . The method of claim 2 , further comprising:
automatically generating, by the at least one processor via a code repository manager, at least one final code that relates to a computer program code based on the at least one final model;
automatically generating, by the at least one processor via a documentation automation manager, at least one model review document that relates to model lifecycle traceability for the at least one final code; and
automatically generating, by the at least one processor via a model explainer, at least one explanation document, the explanation document including behavior information and interaction information that corresponds to at least one from among an input process, an interim process, and an output process of the at least one final model.
5 . The method of claim 4 , wherein the at least one model review document includes at least one from among a data diagnostic information item, a feature engineering information item, a pre-modeling information item, a model optimization information item, a final model information item, and an implementation testing information item.
6 . The method of claim 5 , wherein the implementation testing information item includes user acceptance testing information that relates to at least one from among an alpha test, a beta test, a contract acceptance test, a regulation acceptance test, an operational acceptance test, and a black box test.
7 . The method of claim 1 , further comprising:
examining, by the at least one processor via a model diagnostic tool, a feature distribution for each subset of data that is consumed by the at least one model;
identifying, by the at least one processor via the model diagnostic tool, at least one model instance that performs below a predetermined threshold based on a result of the examining, the at least one model instance corresponding to the at least one model; and
displaying, by the at least one processor via the graphical user interface, diagnostic information for the at least one model, the diagnostic information including feature distribution information and model instance information that corresponds to the identified at least one model instance.
8 . The method of claim 1 , wherein the at least one model includes at least one from among a statistical model, a machine learning model, a mathematical model, a process model, and a data model.
9 . The method of claim 1 , further comprising:
utilizing the chain-style syntax to perform operations asynchronously,
wherein the dynamically adjusting comprises dynamically adjusting without user intervention and from a first processing paradigm to a second processing paradigm, and
wherein at least one from among the first processing paradigm and the second processing paradigm utilizes at least a single processing core.
10 . The method of claim 1 , further comprising:
utilizing a features catalog to automate schema tracking and validate data operations that include the operations involved in producing the processed dataset.
11 . A computing device configured to implement an execution of a method for developing at least one model, the computing device comprising:
a processor;
a memory; and
a communication interface coupled to each of the processor and the memory,
wherein the processor is configured to:
receive, via a graphical user interface, at least one input that relates to workflow that includes at least one from among a data engineering workflow and a feature development workflow,
wherein the data engineering workflow includes a data processing task that identifies an answer to at least one question based on a corresponding user characteristic; and
wherein the feature development workflow includes an iterative software development process that outputs an application based on a predeveloped platform;
determine, via a data assembler, whether data included in the at least one input is in compliance with a predetermined data schema;
generate, via a data explorer, at least one data diagnostic report that relates to an analysis of at least one characteristic for the at least one input;
generate, via the data assembler, at least one modeling strategy that relates to a development of at least one feature for the at least one input;
generate, via the data assembler, the at least one model for the at least one input by using at least one design matrix which comprises a sequential mapping of operations involved in producing a processed dataset and by using multi-stage processor parallelization to determine at least one solution for optimizing the at least one model,
wherein the multi-stage processor parallelization includes separating computation parallelization into multiple processing stages, and
wherein the at least one design matrix relates to at least one explanatory variable from the at least one modeling strategy;
determining, by the at least one processor, a file size of the workflow;
dynamically adjusting, by the at least one processor, a processing paradigm based on the file size,
wherein the processing paradigm relates to a quantity of processing components that are allocated to the workflow,
wherein each of the at least one design matrix includes a collection of at least one feature vector for a plurality of data points, and
wherein each of the plurality of data points correspond to at least one scalar response;
operating the data assembler based on a chain-style syntax that includes arranging operations of the data assembler in a chain; and
utilizing, as input to a subsequent operation of the data assembler, output from a current operation of the data assembler.
12 . The computing device of claim 11 , wherein the processor is further configured to:
automatically optimize, via a data model optimizer, the at least one model by generating at least one final model based on the at least one solution
automatically optimizing, by the at least one processor via a data model optimizer, the at least one model by generating at least one final model based on the at least one solution.
13 . The computing device of claim 12 , wherein, for the optimizing, the processor is further configured to use at least one global optimization routine, the at least one global optimization routine including at least one from among a Bayesian optimization routine, a grid search optimization routine, a reinforcement learning optimization routine, a gradient descent optimization routine, a genetic algorithm optimization routine, a simulated annealing optimization routine, a covariance matrix adaptation evolution strategy (CMA-ES) optimization routine, and a Gaussian process optimization routine.
14 . The computing device of claim 12 , wherein the processor is further configured to:
automatically generate, via a code repository manager, at least one final code that relates to a computer program code based on the at least one final model;
automatically generate, via a documentation automation manager, at least one model review document that relates to model lifecycle traceability for the at least one final code; and
automatically generate, via a model explainer, at least one explanation document, the explanation document including behavior information and interaction information that corresponds to at least one from among an input process, an interim process, and an output process of the at least one final model.
15 . The computing device of claim 14 , wherein the at least one model review document includes at least one from among a data diagnostic information item, a feature engineering information item, a pre-modeling information item, a model optimization information item, a final model information item, and an implementation testing information item.
16 . The computing device of claim 15 , wherein the implementation testing information item includes user acceptance testing information that relates to at least one from among an alpha test, a beta test, a contract acceptance test, a regulation acceptance test, an operational acceptance test, and a black box test.
17 . The computing device of claim 11 , wherein the processor is further configured to:
examine, via a model diagnostic tool, a feature distribution for each subset of data that is consumed by the at least one model;
identify, via the model diagnostic tool, at least one model instance that performs below a predetermined threshold based on a result of the examining, the at least one model instance corresponding to the at least one model; and
display, via the graphical user interface, diagnostic information for the at least one model, the diagnostic information including feature distribution information and model instance information that corresponds to the identified at least one model instance.
18 . The computing device of claim 11 , wherein the at least one model includes at least one from among a statistical model, a machine learning model, a mathematical model, a process model, and a data model.
19 . The computing device of claim 11 , wherein the processor is further configured to:
utilize the chain-style syntax to perform operations asynchronously,
wherein the dynamically adjust comprises dynamically adjusting without user intervention and from a first processing paradigm to a second processing paradigm, and
wherein at least one from among the first processing paradigm and the second processing paradigm utilizes at least a single processing core.
20 . The computing device of claim 11 , wherein the processor is further configured to:
utilize a features catalog to automate schema tracking and validate data operations that include the operations involved in producing the processed dataset.