You can create machine learning (ML) engines based on the ML handlers available in MindsDB. If you can’t find the ML handler of your interest, you can always contribute by building a new ML handler.
Description
The CREATE ML_ENGINE command creates an ML engine that uses one of the available ML handlers.
Syntax
Before creating an ML engine, make sure that the ML handler of your interest is available by querying for the ML handlers.
SELECT *
FROM information_schema.handlers;
SHOW HANDLERS;
If you can’t find the ML handler of your interest, you can contribute by building a new ML handler.
Please note that in the process of contributing new ML engines, ML engines and/or their tests will only run correctly if all dependencies listed in the requirements.txt file are installed beforehand.
If you find the ML handler of your interest, then you can create an ML engine using this command:
CREATE ML_ENGINE [IF NOT EXISTS] ml_engine_name
FROM handler_name
[USING argument_key = argument_value];
Please replace ml_engine_name, handler_name, and optionally, argument_key and argument_value with the real values.
Please do not use the same ml_engine_name as the handler_name to avoid issue while dropping the ML engine.
To verify that your ML engine was successfully created, run the command below:
SELECT *
FROM information_schema.ml_engines;
SHOW ML_ENGINES;
If you want to drop an ML engine, run the command below:
DROP ML_ENGINE ml_engine_name;
Example
Let’s check what ML handlers are currently available:
On execution, we get:
+
| NAME | TITLE | DESCRIPTION | VERSION | CONNECTION_ARGS | IMPORT_SUCCESS | IMPORT_ERROR |
+
| "ray_serve" | "RayServe" | "MindsDB handler for Ray Serve" | "0.0.1" | "[NULL]" | "true" | "[NULL]" |
| "neuralforecast" | "NeuralForecast" | "MindsDB handler for Nixtla's NeuralForecast package" | "0.0.1" | "[NULL]" | "true" | "[NULL]" |
| "autosklearn" | "Auto-Sklearn" | "MindsDB handler for Auto-Sklearn" | "0.0.2" | "[NULL]" | "false" | "No module named 'autosklearn'" |
| "mlflow" | "MLFlow" | "MindsDB handler for MLflow" | "0.0.2" | "[NULL]" | "false" | "No module named 'mlflow'" |
| "openai" | "OpenAI" | "MindsDB handler for OpenAI" | "0.0.1" | "[NULL]" | "true" | "[NULL]" |
| "merlion" | "Merlion" | "MindsDB handler for Merlion" | "0.0.1" | "[NULL]" | "false" | "object.__init__() takes exactly one argument (the instance to initialize)" |
| "byom" | "BYOM" | "MindsDB handler for BYOM" | "0.0.1" | "{'code': {'type': 'path', 'description': 'The path to model code'}, 'modules': {'type': 'path', 'description': 'The path to model requirements'}}" | "true" | "[NULL]" |
| "ludwig" | "Ludwig" | "MindsDB handler for Ludwig AutoML" | "0.0.2" | "[NULL]" | "false" | "No module named 'dask'" |
| "lightwood" | "Lightwood" | "[NULL]" | "1.0.0" | "[NULL]" | "true" | "[NULL]" |
| "huggingface_api" | "Hugging Face API" | "MindsDB handler for Auto-Sklearn" | "0.0.2" | "[NULL]" | "false" | "No module named 'hugging_py_face'" |
| "statsforecast" | "StatsForecast" | "MindsDB handler for Nixtla's StatsForecast package" | "0.0.0" | "[NULL]" | "true" | "[NULL]" |
| "huggingface" | "Hugging Face" | "MindsDB handler for Higging Face" | "0.0.1" | "[NULL]" | "true" | "[NULL]" |
| "TPOT" | "Tpot" | "MindsDB handler for TPOT " | "0.0.2" | "[NULL]" | "false" | "No module named 'tpot'" |
| "langchain" | "LangChain" | "MindsDB handler for LangChain" | "0.0.1" | "[NULL]" | "true" | "[NULL]" |
| "autokeras" | "Autokeras" | "MindsDB handler for Autokeras AutoML" | "0.0.1" | "[NULL]" | "false" | "No module named 'autokeras'" |
+
Here we create an ML engine using the OpenAI handler and providing an OpenAI API key in the USING clause.
CREATE ML_ENGINE my_openai_engine
FROM openai
USING
openai_api_key = '<your opanai api key>';
On execution, we get:
Query successfully completed
Now let’s verify that our ML engine exists.
On execution, we get:
+
|NAME |HANDLER |CONNECTION_DATA |
+
|lightwood |lightwood |{"key":["password"],"value":[""]} |
|huggingface |huggingface |{"key":["password"],"value":[""]} |
|openai |openai |{"key":["password"],"value":[""]} |
|my_openai_engine |openai |{"key":["openai_api_key","password"],"value":["",""]} |
+
Please note that the USING clause is optional, as it depends on the ML handler whether it requires some arguments or not. Here, we created an OpenAI engine and provided own API key.
After creating your ML engine, you can create a model like this:
CREATE MODEL my_model
PREDICT answer
USING
engine = 'my_openai_engine',
prompt_template = 'ask a question to a model'
The USING clause specifies the ML engine to be used for creating a new model.
