Processing natural language based data analytics requests using partially materialized execution plans

A non-transitory computer readable storage medium storing instructions that when executed by one or more computer processors cause the one or more computer processors to perform steps comprising:

• receiving, by an online system, a first natural language data analytics request based on data stored in one or more data stores;
• generating a partially materialized execution plan for processing the first natural language data analytics request using a machine learning based language model, wherein the machine learning based language model non-deterministically generates different outputs when executed multiple times for the same input, wherein the partially materialized execution plan comprises at least a set of instructions identified as materialized, wherein a materialized set of instructions is unchanged if the partially materialized execution plan is regenerated from the first natural language data analytics request using the machine learning based language model;
• iteratively refining the partially materialized execution plan, comprising, repeating one or more times:
• generating a prompt identifying a particular set of instructions and requesting the machine learning based language model to materialize the particular set of instructions by generating an execution plan that uses the particular set of instructions without modification, and
• generating, using the machine learning based language model, an updated partially materialized execution plan that includes the particular set of instructions in addition to one or more sets of instructions materialized during previous iterations;
• storing a finalized execution plan obtained by iteratively refining the partially materialized execution plan;
• executing the finalized execution plan to generate a report;
• receiving a second natural language data analytics request;
• and
• reusing a materialized set of instructions from the execution plan generated for the first natural language data analytics request in an execution plan generated for the second natural language data analytics request.

A computer system comprising:

• one or more computer processors;
• and
• a non-transitory computer readable storage medium storing instructions that when executed by one or more computer processors cause the one or more computer processors to perform steps comprising:
• receiving, by an online system, a first natural language data analytics request based on data stored in one or more data stores;
• generating a partially materialized execution plan for processing the first natural language data analytics request using a machine learning based language model, wherein the machine learning based language model non-deterministically generates different outputs when executed multiple times for the same input, wherein the partially materialized execution plan comprises at least a set of instructions identified as materialized, wherein a materialized set of instructions is unchanged if the partially materialized execution plan is regenerated from the first natural language data analytics request using the machine learning based language model;
• iteratively refining the partially materialized execution plan, comprising,
• repeating one or more times:
• generating a prompt identifying a particular set of instructions and requesting the machine learning based language model to materialize the particular set of instructions by generating an execution plan that uses the particular set of instructions without modification, and
• generating, using the machine learning based language model, an updated partially materialized execution plan that includes the particular set of instructions in addition to one or more sets of instructions materialized during previous iterations;
• storing a finalized execution plan obtained by iteratively refining the partially materialized execution plan;
• executing the finalized execution plan to generate a report;
• receiving a second natural language data analytics request;
• and
• reusing a materialized set of instructions from the execution plan generated for the first natural language data analytics request in an execution plan generated for the second natural language data analytics request.