Natural language query generation for feature stores using zero shot learning
The present disclosure pertains to natural language techniques for querying data stored in feature stores using zero shot learning. In a particular aspect, a computer-implemented method includes receiving a natural language query for retrieving features from a feature store, generating an input prompt by appending a script to the natural language query, and then using a large language model to determine tables or databases from the feature store that are relevant to the natural language query, retrieve metadata for the tables or databases from the feature store, determine feature groups comprising features relevant to the natural language query, and generate a programming language query based on the input prompt, the metadata, and the groups. A list of features within the feature groups that are accessible within the feature store may then be retrieved by executing the programming language query on the feature store.
1. A computer-implemented method comprising:
receiving, from a user, a natural language query for retrieving features from a feature store;
generating an input prompt by appending text identifying the feature store to the natural language query;
determining, by a large language model (LLM), one or more tables or databases from the feature store that are relevant to the natural language query based on the input prompt;
retrieving, by the LLM, metadata for the one or more tables or databases from the feature store;
determining, by the LLM, one or more feature groups comprising features relevant to the natural language query based on the metadata;
generating, by the LLM, a programming language query based on the input prompt, the metadata, and the one or more feature groups;
retrieving a list of features within the one or more feature groups that are accessible within the feature store by executing the programming language query on the feature store; and
outputting the list of features to the user.
2. The computer-implemented method of claim 1 , further comprising:
training, validating, or implementing a machine learning model based on all or a portion of the features in the list of features; or
retrieving a set of data from a database based on the list of features, and training, validating, or implementing a machine learning model using the set of data.
3. The computer-implemented method of claim 1 , wherein the text appended to the input prompt is a script that comprises: identification of the LLM, identification of the feature store, and identification of a generator module that is implemented as part of the LLM and used to orchestrate the extraction of the one or more tables or databases and metadata from the feature store.
4. The computer-implemented method of claim 3 , wherein the generator module comprises an algorithm or series of functional steps that are executed by the LLM to determine the one or more tables or databases from the feature store, retrieve the metadata for the one or more tables or databases from the feature store, and determining the one or more feature groups.
5. The computer-implemented method of claim 4 , wherein executing the algorithm or series of functional steps comprises:
executing a first function that generates and executes, via a query runner function, one or more database queries on one or more databases for retrieval of a list of databases available within the feature store, the list of databases comprising the one or more databases;
executing a second function that generates and executes, via the query runner function, one or more table queries on the one or more databases for retrieval of a list of tables available within the one or more databases, wherein the tables represent feature groups and/or datasets available within the one or more databases and the list of tables comprise the one or more tables; and
executing a third function that generates and executes, via the query runner function, one or more table queries on the tables for retrieval of information about the tables and a list of features within each of the feature groups and/or datasets.
6. The computer-implemented method of claim 5 , wherein the first function also gathers information about the databases, including the databases' structure, the metadata, and schemas, and wherein the third function also gathers information about the tables, including the tables' the metadata and schemas.
7. The computer-implemented method of claim 6 , wherein the programming language query is generated based on the input prompt, the metadata and the schemas for the databases, the metadata and schema for the tables, and the one or more feature groups.
8. A system comprising:
one or more processors; and
one or more computer-readable media storing instructions which, when executed by the one or more processors, cause the system to perform operations comprising:
receiving, from a user, a natural language query for retrieving features from a feature store;
generating an input prompt by appending text identifying the feature store to the natural language query;
determining, by a large language model (LLM), one or more tables or databases from the feature store that are relevant to the natural language query based on the input prompt;
retrieving, by the LLM, metadata for the one or more tables or databases from the feature store;
determining, by the LLM, one or more feature groups comprising features relevant to the natural language query based on the metadata;
generating, by the LLM, a programming language query based on the input prompt, the metadata, and the one or more feature groups;
retrieving a list of features within the one or more feature groups that are accessible within the feature store by executing the programming language query on the feature store; and
outputting the list of features to the user.
9. The system of claim 8 , wherein the operations further comprise:
training, validating, or implementing a machine learning model based on all or a portion of the features in the list of features; or
retrieving a set of data from a database based on the list of features, and training, validating, or implementing a machine learning model using the set of data.
10. The system of claim 8 , wherein the text appended to the input prompt is a script that comprises: identification of the LLM, identification of the feature store, and identification of a generator module that is implemented as part of the LLM and used to orchestrate the extraction of the one or more tables or databases and metadata from the feature store.
11. The system of claim 10 , wherein the generator module comprises an algorithm or series of functional steps that are executed by the LLM to determine the one or more tables or databases from the feature store, retrieve the metadata for the one or more tables or databases from the feature store, and determining the one or more feature groups.
12. The system of claim 11 , wherein executing the algorithm or series of functional steps comprises:
executing a first function that generates and executes, via a query runner function, one or more database queries on one or more databases for retrieval of a list of databases available within the feature store, the list of databases comprising the one or more databases;
executing a second function that generates and executes, via the query runner function, one or more table queries on the one or more databases for retrieval of a list of tables available within the one or more databases, wherein the tables represent feature groups and/or datasets available within the one or more databases and the list of tables comprise the one or more tables; and
executing a third function that generates and executes, via the query runner function, one or more table queries on the tables for retrieval of information about the tables and a list of features within each of the feature groups and/or datasets.
13. The system of claim 12 , wherein the first function also gathers information about the databases, including the databases' structure, the metadata, and schemas, and wherein the third function also gathers information about the tables, including the tables' the metadata and schemas.
14. The system of claim 13 , wherein the programming language query is generated based on the input prompt, the metadata and the schemas for the databases, the metadata and schema for the tables, and the one or more feature groups.
15. One or more non-transitory computer-readable media storing instructions which, when executed by one or more processors, cause a system to perform operations comprising:
receiving, from a user, a natural language query for retrieving features from a feature store;
generating an input prompt by appending text identifying the feature store to the natural language query;
determining, by a large language model (LLM), one or more tables or databases from the feature store that are relevant to the natural language query based on the input prompt;
retrieving, by the LLM, metadata for the one or more tables or databases from the feature store;
determining, by the LLM, one or more feature groups comprising features relevant to the natural language query based on the metadata;
generating, by the LLM, a programming language query based on the input prompt, the metadata, and the one or more feature groups;
retrieving a list of features within the one or more feature groups that are accessible within the feature store by executing the programming language query on the feature store; and
outputting the list of features to the user.
16. The one or more non-transitory computer-readable media of claim 15 , wherein the operations further comprise:
training, validating, or implementing a machine learning model based on all or a portion of the features in the list of features; or
retrieving a set of data from a database based on the list of features, and training, validating, or implementing a machine learning model using the set of data.
17. The one or more non-transitory computer-readable media of claim 16 , wherein the text appended to the input prompt is a script that comprises: identification of the LLM, identification of the feature store, and identification of a generator module that is implemented as part of the LLM and used to orchestrate the extraction of the one or more tables or databases and metadata from the feature store.
18. The one or more non-transitory computer-readable media of claim 17 , wherein the generator module comprises an algorithm or series of functional steps that are executed by the LLM to determine the one or more tables or databases from the feature store, retrieve the metadata for the one or more tables or databases from the feature store, and determining the one or more feature groups.
19. The one or more non-transitory computer-readable media of claim 18 , wherein executing the algorithm or series of functional steps comprises:
executing a first function that generates and executes, via a query runner function, one or more database queries on one or more databases for retrieval of a list of databases available within the feature store, the list of databases comprising the one or more databases;
executing a second function that generates and executes, via the query runner function, one or more table queries on the one or more databases for retrieval of a list of tables available within the one or more databases, wherein the tables represent feature groups and/or datasets available within the one or more databases and the list of tables comprise the one or more tables; and
executing a third function that generates and executes, via the query runner function, one or more table queries on the tables for retrieval of information about the tables and a list of features within each of the feature groups and/or datasets.
20. The one or more non-transitory computer-readable media of claim 19 , wherein the first function also gathers information about the databases, including the databases' structure, the metadata, and schemas, wherein the third function also gathers information about the tables, including the tables' the metadata and schemas, and wherein the programming language query is generated based on the input prompt, the metadata and the schemas for the databases, the metadata and schema for the tables, and the one or more feature groups.