Democratizing AI: A Practical Guide to Snowflake Cortex

Now, let’s use the Cortex FORECAST function to predict sales for the next 7 days for product_id = 1:

SELECT sale_date, sales_amount FROM daily_sales WHERE product_id = 1 ORDER BY sale_date;

SELECT FORECAST(sales_amount, 7) AS predicted_sales, 

 DATEADD(day, seq4(), ‘2023-02-01’) AS predicted_date 

 FROM daily_sales WHERE product_id = 1 AND sale_date = ‘2023-01-31’;

This will give you the forecasted sales for the next 7 days. Cortex supports a fail-fast strategy, letting you experiment and iterate quickly. If your initial models aren’t perfect (and they rarely are!), you can pivot to other methods available through Snowpark and integrate them with Cortex. You might, for example, incorporate external data like weather patterns or promotional campaigns to improve forecast accuracy. You can use more advanced techniques like time series decomposition to identify trends, seasonality, and residual noise in your data, which can further improve forecast accuracy.

Example 2

Let’s look at another example: Sentiment Analysis. This time, we’ll use the SENTIMENT function to analyze customer reviews, but we’ll also explore how to aggregate, contextualize, and visualize the results in order to understand customer perception for your products or services.

Below is the tabular representation of an unstructured NoSQL data

SELECT review_text, SENTIMENT(review_text) AS review_sentiment, product_category

FROM customer_reviews;

SELECT product_category, review_sentiment, COUNT(*) AS review_count

FROM (SELECT review_text, SENTIMENT(review_text) AS review_sentiment, product_category FROM customer_reviews)

    GROUP BY product_category, review_sentiment

    ORDER BY product_category, review_sentiment;

This gives you a count of positive, negative, & neutral reviews for each product category, which you can then visualize in a dashboard or report. Here a simplistic Cortex LLM function allows you identify areas for improvement within specific product lines.