CREATE MODELTYPE

Purpose

Use the CREATE MODELTYPE statement to create a modeltype, which is an object that defines the architecture of ML model to be built by training data. A modeltype can be separately trained on individual dataset, and hyperparameters can be specified for each model training.

Syntax

Diagram

createModeltype

modeltypeSpecClause

Keywords and Parameters

modeltypeName

This is an identifier that specifies the name of the modeltype to be created.

SYNOPSIS

Specify SYNOPSIS if the modeltype is a synopsis generative modeltype.

INFERENCE

Specify INFERENCE if the modeltype is a inferential modeltype.

modeltypeSpecClause

This clause specifies the details of the modeltype.

LOCAL

Specify LOCAL if the modeltype is in your local machine.

REMOTE

Specify REMOTE if the modeltype is in a remote machine.

‘modeltypeClassName’

This is a string literal that specifies the class of the model type in the file/service at modelUri.

‘modelUri’

This is a string literal that specifies the directory path for local modeltype or the connection URL for remote modeltype.

Examples

Creating a Local Modeltype

The following statement creates a local synopsis generative modeltype tablegan as the class TableGAN in the directory path ‘models/TableGAN.py’.

CREATE MODELTYPE tablegan FOR SYNOPSIS AS LOCAL CLASS 'TableGAN' IN 'models/TableGAN.py';

The following statement creates a local inferential modeltype rspn as the class RSPN in the directory path ‘models/RSPN.py’.

CREATE MODELTYPE rspn FOR INFERENCE AS LOCAL CLASS 'RSPN' IN 'models/RSPN.py';

Creating a Remote Modeltype

The following statement creates a remote synopsis generative modeltype remote_tablegan as the class TableGAN in the directory path ‘http://<host>:<port>’. It is assumed that the TrainDB model server is running at the specified address.

CREATE MODELTYPE remote_tablegan FOR SYNOPSIS AS REMOTE CLASS 'TableGAN' IN 'http:/<host>:<port>/';