MindsDB NestJS Module

A NestJS module for integrating with MindsDB. This allows for easy model management and prediction within your NestJS application.

Installation

# npm
npm install @oak-digital/nestjs-mindsdb
# pnpm
pnpm install @oak-digital/nestjs-mindsdb

Setup

This module works by defining your models in your code. It is currently not possible to create models dynamically.

In your module, you should import this module with the forRoot method.

// app.module.ts
@Module({
  imports: [
    MindsdbModule.forRoot({
      // The predefined models to use
      models: myModels,
      // The default project to use, 'mindsdb' by default
      project: 'mindsdb',
      // If the connection is omitted,
      // it will use the following default connection options
      connection: {
        host: 'http://127.0.0.1:47334',
        user: 'mindsdb',
        password: '',
        managed: false,
      },
    }),
  ],
})

Defining Models

Models in MindsDB are described through the IModel interface. Here's how you can define your models:

import { Granularity, IModel } from "./mindsdb.models";

export const Models = new Map<string, IModel>([
  [
    "balance_auto",
    {
      name: "balance_auto",
      granularity: Granularity.day,
      targetColumn: "sum",
      predictOptions: {
        join: "views.enriched_balance",
        where: ["t.customerId = $CUSTOMER_ID$", "t.date > $DATE$"],
      },
      trainingOptions: {
        select: "select * from enriched_balance",
        groupBy: "customerId",
        orderBy: "date",
        horizon: 60,
        window: 90,
        // using: {
        //   submodels: [{ module: 'GluonTSMixer', args: {} }],
        // },
      },
      finetuneOptions: {
        select: "select * from enriched_balance where customerId = $CUSTOMER_ID$ and date > $DATE$",
        integration: undefined,
      },
    },
  ],
  [
    "balance_gluon",
    {
      name: "balance_gluon",
      granularity: Granularity.day,
      targetColumn: "sum",
      view: {
        select: "select * from production.banking.enriched_balance",
        name: "balance_gluon_enriched_balance",
      },
      predictOptions: {
        join: "views.enriched_balance",
        where: ["t.customerId = $CUSTOMER_ID$", "t.date > $DATE$"],
      },
      trainingOptions: {
        select: "select * from balance_gluon_enriched_balance",
        groupBy: "customerId",
        orderBy: "date",
        horizon: 60,
        window: 90,
        // using: {
        //   submodels: [{ module: 'GluonTSMixer', args: {} }],
        // },
      },
      finetuneOptions: {
        select: "select * from balance_gluon_enriched_balance",
        integration: undefined,
      },
    },
  ],
]);

Placeholders such as $CUSTOMER_ID$ and $DATE$ can be used in the model definition. These will be replaced when you're making predictions.

Usage

Using the MindsdbService

The MindsdbService provides several methods that correspond to different operations you can perform on your models. If parameters are not provided will use the one's defined on the IModel object for each.

1. Create a Model

The models are not created by default, so you need to create them with the create method

const createDto = new CreateMindsdbDto();
createDto.name = 'balance_auto';
mindsdbService.create(createDto);

2. Predict Using a Model

const predictDto = new PredictMindsdbDto();
predictDto.join = "views.enriched_balance";
predictDto.where = ["t.customerId = $CUSTOMER_ID$", "t.date > $DATE$"];
predictDto.limit = 100;
predictDto.params = {
  $CUSTOMER_ID$: 1234,
  $DATE$: new Date('2023-01-01')
};
mindsdbService.predict('balance_auto', predictDto);

3. Fine-tune a Model

const finetuneDto = new FinetuneMindsdbDto();
finetuneDto.select = "select * from enriched_balance where customerId = $CUSTOMER_ID$ and date > $DATE$";
finetuneDto.params = {
  $CUSTOMER_ID$: 1234,
  $DATE$: new Date('2023-01-01')
};
mindsdbService.finetune('balance_auto', finetuneDto);

4. Retrain a Model

const retrainDto = new RetrainMindsDbDto();
retrainDto.select = "select * from balance_gluon_enriched_balance";
retrainDto.groupBy = "customerId";
retrainDto.orderBy = "date";
retrainDto.window = 90;
retrainDto.horizon = 60;
mindsdbService.retrain('balance_auto', retrainDto);

5. Retrieve All Models

const allModels = mindsdbService.findAll();

6. Retrieve a Specific Model

mindsdbService.findOne('balance_auto');

7. Remove a Model

mindsdbService.remove('balance_auto');

Setting up the Controller

To expose your models through a RESTful API, extend the AbstractMindsdbController and use NestJS's @Controller decorator. This controller will automatically have all the endpoints corresponding to the methods in the AbstractMindsdbController.

import { Controller } from '@nestjs/common';
import { MindsdbService } from '@oak-digital/nestjs-mindsdb';
import { AbstractMindsdbController } from '@oak-digital/nestjs-mindsdb';

@Controller('mindsdb')
export class MindsdbController extends AbstractMindsdbController {
  constructor(mindsdbService: MindsdbService) {
    super(mindsdbService);
  }
}

Remember to adjust any paths, references, and examples according to your project structure and requirements.