System and method for heterogeneous model composition

1. A heterogeneous model composition system, comprising:

a non-transitory computer readable medium; and

a processor coupled to the non-transitory computer readable medium, the processor configured to:

determine a series of heterogeneous models;

receive a requested output of an intermediary model of the series;

determine a subset of models connecting a root model of the series of heterogeneous models to the intermediary model, the subset of models comprising model layers;

identify a subset of the model layers connecting the root model to the intermediary model;

convert input data into a standard input object;

facilitate execution of the subset of model layers, comprising, for each successive model within the subset of models:

a) converting the standard input object into a model-specific input (MSI) object specific to the model within the subset of models;

b) executing the model within the subset of models using the MSI object to generate a model-specific output (MSO) object;

c) converting the MSO object to a standard output object; and

d) setting the standard output object from the model within the subset of models as the standard input object for a successive model;

convert the standard output object of the intermediary model to an output datatype; and

return the standard output object from the intermediary model to an endpoint.

2. The system of claim 1 , wherein the series of heterogeneous models is selected by a user using an interface.

3. The system of claim 1 , wherein the series of heterogeneous models is represented as a directed acyclic graph.

4. The system of claim 1 , wherein each model is associated with a pre-processor configured to convert a standard input object to a MSI object and a post-processor configured to convert a MSO object to a standard output object, wherein the pre-processor and the post-processor are specific to the model.

5. The system of claim 4 , wherein a post-processor of a preceding model within the series is connected to a pre-processor of a successive model within the series.

6. The method of claim 1 , wherein the endpoint comprises at least one of: a user interface executing on a device, a database, or an API.

7. A heterogeneous model composition system, comprising:

a deserializer configured to deserialize an input into a standard format;

a model composition comprising a series of heterogeneous models, each model comprising:

a pre-processor configured to convert an input for the model from the standard format to a model-specific input (MSI) format; and

a post-processor configured to convert an output for the model from a model-specific output (MSO) format to the standard format;

a serializer configured to serialize an output of a model from the series into an output datatype; and

an execution system configured to:

receive a requested output for an intermediary model of the series;

determine model layers of a subset of models, the model layers connecting a root model of the model composition to the intermediary model; and

facilitate execution of the model layers of the subset of models, comprising:

converting an input into a standard input object using the deserializer;

for each successive model within the subset of models:

determining a model-specific input (MSI) object from the standard input object using the pre-processor for the model within the subset of models;

executing the model within the subset of models using the MSI object to generate a model-specific output (MSO) object;

determining a standard output object from the MSO object using the post-processor for the model within the subset of models; and

using the standard output object from the model within the subset of models as the standard input object for a successive model; and

converting the standard output object of the intermediary model into an output datatype using the serializer.

8. The system of claim 7 , wherein each model within the series is authored by a different entity.

9. The system of claim 7 , wherein a post-processor of a preceding model within the series is connected to a pre-processor of a successive model within the series.

10. The system of claim 7 , wherein models within the series comprise independently trained machine learning models.

11. The system of claim 7 , wherein the model composition is determined by a user at an interface.

12. The system of claim 7 , wherein the deserializer is connected to the root model and the serializer is connected to the intermediary model.

13. A method comprising:

determining a set of connected heterogeneous models comprising a parent model and a set of child models;

receiving a requested output for a child model of the set of child models;

identifying model layers of a subset of models between the parent model and the child model;

receiving an input;

converting the input into a standard format;

for each successive child model within the subset of models:

a) converting the standard-formatted input into a model-specific input (MSI) format for the child model within the subset of models using a pre-processor that only accepts an input in the standard format;

b) executing the child model within the subset of models using the MSI-formatted input to generate an output in a model-specific output (MSO) format for the child model within the subset of models; and

c) converting the output from the MSO format to the standard format; and

wherein a)-c) are performed using the identified model layers, wherein a)-c) are repeated for successive child models within the subset of models, using the standard-formatted output from a prior model as the standard-formatted input for the successive child model;

converting the standard-formatted output of a child model to an output datatype; and

providing the output datatype to an endpoint.

14. The method of claim 13 , wherein heterogeneous models of the set are trained contemporaneously.

15. The method of claim 13 , wherein the set of heterogeneous models is selected by a user using an interface.

16. The method of claim 13 , wherein each model within the set is authored by a different entity.