Skip to content
RelationalAI Private Docs

Using the RelationalAI Predictive Reasoner in a Snowflake Notebook

Get Access to the RelationalAI Native App

Section titled “Get Access to the RelationalAI Native App”

To use the RelationalAI Predictive Reasoner, you’ll need access to the RelationalAI Native App. You can request access via the Snowflake Marketplace. When submitting your request, be sure to include a note indicating that you’re interested in the app’s predictive reasoning capabilities using graph neural networks (GNNs).

Once your request is approved you will receive email notification by RAI. Follow the instructions here to install the RAI Native App. Note that you will need to be a user with either ORGADMIN or ACCOUNTADMIN privileges to do the installation.

If you have already installed the RelationalAI Native App reach out to us to request enabling the predictive reasoning feature flag.

Set up Notebook

Section titled “Set up Notebook”

  1. Create Notebook

    • Go to https://app.snowflake.com.
    • Navigate to Projects > Notebooks and click + Notebook.
    • Enter a Name for the notebook.
    • In Notebook location, choose the database and schema where the notebook will be stored.
    • In Runtime, select Run on container.
    • In Runtime version, choose GPU for faster model training. Otherwise, choose CPU.
    • Select a Compute pool, which will run the Python commands in the notebook.
    • Select a Query warehouse, which will execute the SQL queries from the notebook.

  2. Configure External Access

    Click the menu button in the top-right corner and select Notebook settings. In the External access section, enable PYPI_ACCESS_INTEGRATION and click Save. This allows the notebook to use pip to install external Python packages from PyPI.

  3. Install the relationalai_gnns Package

    Click Start to launch the notebook kernel. The relationalai_gnns Python package is required to interact with the RelationalAI Native App. In the first cell of the notebook, run:

     !pip install relationalai-gnns

Install Graphviz for Schema Diagrams

Section titled “Install Graphviz for Schema Diagrams”

.visualize_dataset() renders diagrams using the Graphviz dot binary, which isn’t included in the notebook’s container by default. To enable it:

  1. Create a network rule and external access integration for the Ubuntu package mirrors:

    CREATE OR REPLACE NETWORK RULE apt_network_rule
      MODE = EGRESS
      TYPE = HOST_PORT
      VALUE_LIST = (
        'archive.ubuntu.com:80', 'archive.ubuntu.com:443',
        'security.ubuntu.com:80', 'security.ubuntu.com:443',
        'ports.ubuntu.com:80', 'ports.ubuntu.com:443'
      );
    

    CREATE OR REPLACE EXTERNAL ACCESS INTEGRATION apt_access_integration
      ALLOWED_NETWORK_RULES = (apt_network_rule)
      ENABLED = TRUE;
    

    GRANT USAGE ON INTEGRATION apt_access_integration TO ROLE <ROLE>;

    Grant USAGE to whichever role the notebook actually runs as — check with DESC NOTEBOOK <DATABASE>.<SCHEMA>.<NOTEBOOK_NAME> if you’re not sure.

  2. Attach the integration to the notebook:

    ALTER NOTEBOOK <DATABASE>.<SCHEMA>.<NOTEBOOK_NAME>
      SET EXTERNAL_ACCESS_INTEGRATIONS = (apt_access_integration);

  3. Install graphviz from a Python cell:

    !apt-get update && apt-get install -y graphviz