Create, Train, and Deploy a Model

Description

The CREATE MODEL statement creates and trains a machine learning (ML) model.

Please note that the CREATE MODEL statement is equivalent to the CREATE PREDICTOR statement. We are transitioning to the CREATE MODEL statement, but the CREATE PREDICTOR statement still works.

Syntax

Overview

Here is the full syntax:

CREATE [OR REPLACE] MODEL [IF NOT EXISTS] project_name.predictor_name

[FROM [integration_name | project_name]
    (SELECT [sequential_column,] [partition_column,] column_name, ...
     FROM [integration_name. | project_name.]table_name
     [JOIN model_name])]

PREDICT target_column

[ORDER BY sequential_column]
[GROUP BY partition_column]
[WINDOW int]
[HORIZON int]

[USING engine = 'engine_name',
       tag = 'tag_name',
       ...];

Where:

Expressions	Description
`project_name`	Name of the project where the model is created. By default, the `mindsdb` project is used.
`predictor_name`	Name of the model to be created.
`integration_name`	Name of the integration created using the `CREATE DATABASE` statement or file upload.
`(SELECT column_name, ... FROM table_name)`	Selecting data to be used for training and validation.
`target_column`	Column to be predicted.
`ORDER BY sequential_column`	Used in time series models. The column by which time series is ordered. It can be a date or anything that defines the sequence of events.
`GROUP BY partition_column`	Used in time series models. It is optional. The column by which rows that make a partition are grouped. For example, if you want to forecast the inventory for all items in the store, you can partition the data by `product_id`, so each distinct `product_id` has its own time series.
`WINDOW int`	Used in time series models. The number of rows to look back at when making a prediction. It comes after the rows are ordered by the column defined in `ORDER BY` and split into groups by the column(s) defined in `GROUP BY`. The `WINDOW 10` syntax could be interpreted as “Always use the previous 10 rows”.
`HORIZON int`	Used in time series models. It is optional. It defines the number of future predictions (it is 1 by default). However, the `HORIZON` parameter, besides defining the number of predictions, has an impact on the training procedure when using the Lightwood ML backend. For example, different mixers are selected depending on whether the `HORIZON` value is one or greater than one.
`engine_name`	You can optionally provide an ML engine, based on which the model is created.
`tag_name`	You can optionally provide a tag that is visible in the `training_options` column of the `mindsdb.models` table.

Regression Models

Here is the syntax for regression models:

CREATE MODEL project_name.predictor_name
FROM integration_name
    (SELECT column_name, ... FROM table_name)
PREDICT target_column
[USING engine = 'engine_name',
       tag = 'tag_name'];

Please note that the FROM clause is mandatory here.

The target_column that will be predicted is a numerical value. The prediction values are not limited to a defined set of values, but can be any number from the given range of numbers.

Classification Models

Here is the syntax for classification models:

CREATE MODEL project_name.predictor_name
FROM integration_name
    (SELECT column_name, ... FROM table_name)
PREDICT target_column
[USING engine = 'engine_name',
       tag = 'tag_name'];

Please note that the FROM clause is mandatory here.

The target_column that will be predicted is a string value. The prediction values are limited to a defined set of values, such as Yes and No.

Time Series Models

Here is the syntax for time series models:

CREATE MODEL project_name.predictor_name
FROM integration_name
    (SELECT sequential_column, partition_column,
            other_column, target_column
     FROM table_name)
PREDICT target_column

ORDER BY sequential_column
[GROUP BY partition_column]

WINDOW int
[HORIZON int]

[USING engine = 'engine_name',
       tag = 'tag_name'];

Please note that the FROM clause is mandatory here.

Due to the nature of time series forecasting, you need to use the JOIN statement and join the data table with the model table to get predictions.

NLP Models

Here is the syntax for using external models within MindsDB:

CREATE MODEL project_name.model_name
PREDICT PRED
USING
  engine = 'engine_name',
  task = 'task_name',
  model_name = 'hub_model_name',
  input_column = 'input_column_name',
  labels = ['label1', 'label2'];

Please note that you don’t need to define the FROM clause here. Instead, the input_column is defined in the USING clause.

It allows you to bring an external model, for example, from the Hugging Face model hub, and use it within MindsDB.

Large Language Models (LLM)

MindsDB integrates with numerous LLM providers listed here.

Commonly, LLMs support the prompt_template parameter that stores the message/instruction to the model.

CREATE MODEL project_name.model_name
PREDICT answer
USING
  engine = 'llm_engine_name',
  prompt_template = 'answer users questions in a helpful way: {{questions}}';

The prompt_template parameter instructs the model what output should be generated. It can include variables enclosed in double curly braces, which will be replaced with data values upon joining the model with the input data.

Example

Regression Models

Here is an example for regression models that uses data from a database:

CREATE MODEL mindsdb.home_rentals_model
FROM example_db
  (SELECT * FROM demo_data.home_rentals)
PREDICT rental_price
USING engine = 'lightwood',
      tag = 'my home rentals model';

On execution, we get:

Query OK, 0 rows affected (x.xxx sec)

Visit our tutorial on regression models to see the full example.

Classification Models

Here is an example for classification models that uses data from a file:

CREATE MODEL mindsdb.customer_churn_predictor
FROM files
  (SELECT * FROM churn)
PREDICT Churn
USING engine = 'lightwood',
      tag = 'my customers model';

On execution, we get:

Query OK, 0 rows affected (x.xxx sec)

Visit our tutorial on classification models to see the full example.

Time Series Models

Here is an example for time series models that uses data from a file:

CREATE MODEL mindsdb.house_sales_predictor
FROM files
  (SELECT * FROM house_sales)
PREDICT MA
ORDER BY saledate
GROUP BY bedrooms
-- the target column to be predicted stores one row per quarter
WINDOW 8      -- using data from the last two years to make forecasts (last 8 rows)
HORIZON 4;    -- making forecasts for the next year (next 4 rows)

On execution, we get:

Query OK, 0 rows affected (x.xxx sec)

Visit our tutorial on time series models to see the full example.

NLP Models

Here is an example for the Hugging Face model:

CREATE MODEL mindsdb.spam_classifier
PREDICT PRED
USING
  engine = 'huggingface',
  task = 'text-classification',
  model_name = 'mrm8488/bert-tiny-finetuned-sms-spam-detection',
  input_column = 'text_spammy',
  labels = ['ham', 'spam'];

On execution, we get:

Query OK, 0 rows affected (x.xxx sec)

Large Language Models (LLM)

Here is an example using the OpenAI engine:

CREATE MODEL sentiment_classifier
PREDICT sentiment
USING
  engine = 'openai_engine',
  prompt_template = 'analyze customer reviews and assign sentiment as positive or negative or neutral: {{review}}';

Note that the prompt_template parameter stores instructions that the model will follow to generate output.

Visit our page no how to bring Hugging Face models into MindsDB for more details.

Checking Model Status

After you run the CREATE MODEL statement, you can check the status of the training process by querying the mindsdb.models table.

DESCRIBE predictor_name;

On execution, we get:

+---------------+-----------------------------------+----------------------------------------+-----------------------+-------------+---------------+-----+-----------------+----------------+
|name           |status                             |accuracy                                |predict                |update_status|mindsdb_version|error|select_data_query|training_options|
+---------------+-----------------------------------+----------------------------------------+-----------------------+-------------+---------------+-----+-----------------+----------------+
|predictor_name |generating or training or complete |number depending on the accuracy metric |column_to_be_predicted |up_to_date   |22.7.5.0       |     |                 |                |
+---------------+-----------------------------------+----------------------------------------+-----------------------+-------------+---------------+-----+-----------------+----------------+

SQL API

​Description

​Syntax

​Overview

​Regression Models

​Classification Models

​Time Series Models

​NLP Models

​Large Language Models (LLM)

​Example

​Regression Models

​Classification Models

​Time Series Models

​NLP Models

​Large Language Models (LLM)

Description

Syntax

Overview

Regression Models

Classification Models

Time Series Models

NLP Models

Large Language Models (LLM)

Example

Regression Models

Classification Models

Time Series Models

NLP Models

Large Language Models (LLM)