Use the grant SELECT access on in-database models for inferencing. For a specific database user, use:

GRANT SELECT ON MINING MODEL <model name> TO
    username;

You can obtain the Oracle Machine Learning User Management URL for a specific tenancy from the Oracle Cloud Infrastructure (OCI) command line. You must first obtain the tenancy ID.

1. To obtain the tenancy ID, go to your OCI Profile on the top right corner of the Oracle Cloud page and click Tenancy.
2. On the Tenancy details page, click Copy to obtain the tenancy URL.
3. Type the following command in your OCI command line interface:

   oci db database list  --compartment-id  <tenancy OCID>

   Here,

   • compartment-id: This is the unique ID assigned to your compartment.
   • OCID: This is the Oracle Cloud Identifier (OCID) for your tenancy.
4. This command returns the following:

   "connection-urls": {
           "apex-url": https://<tenancy ID>-<database name>.<region>.oraclecloudapps.com/ords/apex,
           "graph-studio-url": https://<tenancy ID>-<database name>.<region>.oraclecloudapps.com/graphstudio/,
           "machine-learning-user-management-url": https://<tenancy ID>-<database name>.<region>-1.oraclecloudapps.com/omlusers/,
           "sql-dev-web-url": https://<tenancy ID>-<database name>.<region>-1.oraclecloudapps.com/ords/sql-developer
         },

Use the view v$pdbs and run the following:

select 'https://' || 
        lower(REPLACE(p.name, '_', '-')) ||
        '.' ||
        REGEXP_REPLACE(j.PUBLIC_DOMAIN_NAME, '[^.]+', 'oraclecloudapps', 1, 3) ||
        '/omlusers/' auth_token_url,
        'https://' || 
        lower(REPLACE(p.name, '_', '-')) ||
        '.' ||
        REGEXP_REPLACE(j.PUBLIC_DOMAIN_NAME, '[^.]+', 'oraclecloudapps', 1, 3) ||
        '/oml/' embed_python_r_url,
        'https://' || 
        lower(REPLACE(p.name, '_', '-')) ||
        '.' ||
        REGEXP_REPLACE(j.PUBLIC_DOMAIN_NAME, '[^.]+', 'oraclecloudapps', 1, 3) ||
        '/omlmod/' rest_services_url
from v$pdbs p,
     json_table(p.cloud_identity, '$' columns (
        PUBLIC_DOMAIN_NAME path '$.PUBLIC_DOMAIN_NAME'                                      
     )) j;

This query returns three URLs:

• OML User Management URL (same as returned by OCI CLI)
• URL for OML4Py and OML4R embedded execution
• URL for OML Services REST APIs

You can see the URLs for your database in Database Actions

1. On your Oracle Autonomous AI Database console, click Database actions, and then select the option View all database actions.
2. Click RESTFul Services and then click Oracle Machine Learning RESTful services. The Oracle Machine Learning RESTful Services dialog opens.
3. On the Oracle Machine Learning RESTful Services dialog, copy the URL for your ADB instance.

For this, use the materialize() function. This function pushes the contents represented by an OML proxy object into a table in the database.

proxy_object.materialize(table="MYTABLENAME")

For a list of overloaded transparency layer functions, run:

%python

import pandas as pd
import oml

# Create a simple Pandas DataFrame
df = pd.DataFrame({
    'col1': [1, 2, 3],
    'col2': ['a', 'b', 'c']})

# Create an OML proxy object
oml_df = oml.create(df, "df")

# List all public methods and attributes
methods = [m for m in oml_df.__dir__() if not m.startswith('_')]
methods.sort()
print(methods)

For Oracle Machine Learning solutions deployment, every UDF has its own oml.connect(). Instead of hard-coding credentials into the oml.connect statement, is there an alternative solution that can work on multiple database environments?

• For the Python API for embedded execution, use oml_connect=True
• For SQL API, use "oml_connect":1
• For more information, see pyqTableEval Function (On-Premises Database)

Besides OML's in-database models, you can also host your own models on Oracle Machine Learning.

• Native Python and R models can be stored using the OML4Py and OML4R datastores. You can call these models through Python, R, SQL, and REST endpoints, optionally in parallel.
• ONNX format classification, regression, feature extraction, and clustering models can be imported and use the same SQL prediction operators as native models. You can also deploy these models to OML Services for real-time scoring through REST endpoints.
• Transformer models from Hugging Face can be converted to ONNX format, augmented through a pipeline, and loaded to the database for use with AI Vector Search.
• LLMs can be accessed through Select AI for SQL generation, RAG, and chat.

For more information, see:

• Enhance your AI/ML applications with flexible Bring Your Own Model options
• Bring Your Own Model to Your Database for AI and ML

Oracle Machine Learning does not directly consume data from OCI Streaming or Kafka since it works with database tables. But there are several ways to get streaming data into tables for OML to use. You could use:

• OCI GoldenGate for real-time sync to Oracle Autonomous AI Database
• OCI Connector Hub along with OCI Functions to process and load stream data
• OCI Functions triggered by streaming events

This is the workflow:

1. Use data from OCI Streaming
2. Load data into Oracle Autonomous AI Database tables
3. Use Oracle Machine Learning to build models on those tables using Oracle Machine Learning for SQL, Oracle Machine Learning for R, or Oracle Machine Learning for Python

%conda

create -n mypyenv -c conda-forge --override-channels --strict-channel-priority python=3.13.5 seaborn

upload mypyenv --overwrite -t application "OML4PY"

To determine the python version being used, run the following command:

import sys
sys.version

In this command:

• -n: This is the name of the environment. In this example, it is mypyenv.
• -c: This is the channel name. In this example, the conda-forgechannel is used to install the Python package.
• --override-channels: This argument ensures that the system does not search default, and requires a channel to be mentioned.
• --strict-channel-priority: This argument ensures that packages in lower priority channels are not considered if a package with the same name appears in a higher priority channel. In this example, the priority is given to Python 3.13.5 and seaborn.
• upload: This argument uploads the Conda environment named mypyenv to a target location.
• --overwrite: This argument ensures that if a file or directory with the same name already exists at the target location, it should be overwritten.
• -t application: This argument specifies the type of the target as an application.
• "OML4PY": This is the name or path of the target application where the environment will be uploaded.

%conda

create -n myrenv -c conda-forge --override-channels --strict-channel-priority r-base=4.0.5 r-forecast

upload myrenv --overwrite -t application "OML4R"

In this command:

• -n: This is the name of the environment. In this example, it is myrenv.
• -c: This is the channel name. In this example, it is conda-forge.
• --override-channels: This argument ensures that the system does not search default, and requires a channel to be mentioned.
• --strict-channel-priority: This argument ensures that packages in lower priority channels are not considered if a package with the same name appears in a higher priority channel. In this example, the priority is given to R-4.0.5 and forecast.
• upload: This argument uploads the Conda environment named myrenv to a target location.
• --overwrite: This argument ensures that if a file or directory with the same name already exists at the target location, it should be overwritten.
• -t application: This argument specifies the type of the target as an application.
• "OML4R": This is the name or path of the target application where the environment will be uploaded.