Process & Analyze Hugging Face Data in Databricks (AWS)

Databricks is a cloud-based service that provides data processing capabilities through Apache Spark. When paired with the CData JDBC Driver, customers can use Databricks to perform data engineering and data science on live Hugging Face data. This article explains how to host the CData JDBC Driver in AWS, as well as connect to and process live Hugging Face data in Databricks.

With built-in optimized data processing, the CData JDBC Driver offers unmatched performance for interacting with live Hugging Face data. When you issue complex SQL queries to Hugging Face, the driver pushes supported SQL operations, like filters and aggregations, directly to Hugging Face and utilizes the embedded SQL engine to process unsupported operations client-side (often SQL functions and JOIN operations). Its built-in dynamic metadata querying allows you to work with and analyze Hugging Face data using native data types.

Install the CData JDBC Driver in Databricks

To work with live Hugging Face data in Databricks, install the driver on your Databricks cluster.

1. Navigate to your Databricks administration screen and select the target cluster.
2. On the Libraries tab, click "Install New."
3. Select "Upload" as the Library Source and "Jar" as the Library Type.
4. Upload the JDBC JAR file (cdata.jdbc.api.jar) from the installation location (typically C:\Program Files\CData[product_name]\lib).

Access Hugging Face Data in your Notebook: Python

With the JAR file installed, we are ready to work with live Hugging Face data in Databricks. Start by creating a new notebook in your workspace. Name the notebook, select Python as the language (though Scala is available as well), and choose the cluster where you installed the JDBC driver. When the notebook launches, we can configure the connection, query Hugging Face, and create a basic report.

Configure the Connection to Hugging Face

Connect to Hugging Face by referencing the JDBC Driver class and constructing a connection string to use in the JDBC URL. Additionally, you will need to set the RTK property in the JDBC URL (unless you are using a Beta driver). You can view the licensing file included in the installation for information on how to set this property.

Step 1: Connection Information

driver = "cdata.jdbc.api.APIDriver"
url = "jdbc:api:RTK=5246...;Profile=C:\profiles\HuggingFace.apip;ProfileSettings='APIKey=hf_xxxxxxxxxxxxxxxxxxxx';"

Built-in Connection String Designer

For assistance in constructing the JDBC URL, use the connection string designer built into the Hugging Face JDBC Driver. Either double-click the JAR file or execute the jar file from the command-line.

java -jar cdata.jdbc.api.jar

Fill in the connection properties and copy the connection string to the clipboard.

HuggingFace Hub uses token-based authentication to enable access to its API. The API provides access to machine learning models, datasets, spaces, papers, and other resources on the HuggingFace Hub platform.

Using API Key Authentication

To authenticate to HuggingFace Hub, you will need to provide an API Key (Access Token). To obtain your access token:

1. Log in to your HuggingFace account at https://huggingface.co
2. Navigate to Settings > Access Tokens
3. Click "New token" to create a new access token
4. Select the appropriate permissions (read or write)
5. Copy the token value

After obtaining your access token, set the following connection properties:

• AuthScheme: Set this to APIKey.
• APIKey: Set this to your HuggingFace access token.

Example connection string

Profile=C:\profiles\HuggingFace.apip;ProfileSettings='APIKey=hf_xxxxxxxxxxxxxxxxxxxx';

Load Hugging Face Data

Once you configure the connection, you can load Hugging Face data as a dataframe using the CData JDBC Driver and the connection information.

Step 2: Reading the data

remote_table = spark.read.format ( "jdbc" ) \
	.option ( "driver" , driver) \
	.option ( "url" , url) \
	.option ( "dbtable" , "Collections") \
	.load ()

Display Hugging Face Data

Check the loaded Hugging Face data by calling the display function.

Step 3: Checking the result

display (remote_table.select (""))

Analyze Hugging Face Data in Databricks

If you want to process data with Databricks SparkSQL, register the loaded data as a Temp View.

Step 4: Create a view or table

remote_table.createOrReplaceTempView ( "SAMPLE_VIEW" )

With the Temp View created, you can use SparkSQL to retrieve the Hugging Face data for reporting, visualization, and analysis.

% sql

SELECT ,  FROM SAMPLE_VIEW ORDER BY  DESC LIMIT 5

The data from Hugging Face is only available in the target notebook. If you want to use it with other users, save it as a table.

remote_table.write.format ( "parquet" ) .saveAsTable ( "SAMPLE_TABLE" )

Download a free, 30-day trial of the CData API Driver for JDBC and start working with your live Hugging Face data in Databricks. Reach out to our Support Team if you have any questions.