Integrate with Snowflake

Duet offers a number of direct integrations to get Duet published models to process your textual data where they are stored and produce the model output in the same location as your data. These integrations enable users to have a full no-code experience from building a model to consuming the model on their data where it is stored.

1. From the Publish tab, publish your latest changes. Once you finish publishing, press the Snowflake icon.
   
   This will open a Snowflake login dialog. You must have a pre-existing Snowflake account. To create one, visit Snowflake. Snowflake works with all cloud providers.
   

2. To obtain your Snowflake account credentials, locate your Username and Password on your Snowflake account from the Profile page. You can access your Account ID by clicking the Accounts tab under the Admin header in the left side-bar on your Snowflake account. Snowflake has more information about Account IDs in their documentation, but you can locate it easily by clicking the copy link pictured below, and removing "https://" from the front and ".snowflakecomputing.com" from the end. In other words, the Account ID is the company identifier (FCYPURH) and your personal account identifier (mpb29816). They are combined in lowercase with a "-" between them.
   

3. To start viewing your Snowflake databases and tables in Duet, press "Start New Connection". You can have a max number of 10 connections.

4. Give your connection a unique name. Then specify the warehouse, database, schema, and database table where your textual data that the Duet model will process resides. Your table must have a primary key column. Click "Load Table Headers".
   

5. Select the input column that has your textual data.
   
   Once selected, it will be highlighted around the border like this.
   

6. Select the Duet model deployment version you want to run on your textual data. To see the output of your processed data, you can either select an existing column (which will then be overridden) or you can add a new column to your existing table. The output column pictured below is a new one. Specify the name of the new column you want the Duet model predictions to be stored in. Press "Enter" to save the name. Be sure to press "Save" at the bottom right of the screen. You can always edit the settings of your saved connections.
   

7. Now your Snowflake job is ready to run. Click "Run". The Snowflake job might take time if your textual data is large. You can track the progress of your job or cancel it under Active Jobs. Please note that you can log out of Duet and the job will continue to run; it will only stop if you explicitly cancel it. You have a maximum of 5 concurrent jobs.
   
   Once completed, it will move to Past Jobs.
   

8. You can check back in on Snowflake to see The Duet model prediction column. For a classifier, the prediction column would show the top predicted category in the output column. For an entity extractor, the detected top level entities will show in the output column. If multiple instances of the same entity type are detected, they will display as comma separated. For example, if you are storing the detected US addresses and one of the documents has two addresses, the output column will show "123 main street Bellevue WA 98004", “234 Avondale street Redmond WA 98052”.