The goal of this walkthrough is to provide SQL programmers with hands-on experience building a machine learning solution in SQL Server. In this walkthrough, you'll learn how to incorporate Python into an application by adding Python code to stored procedures.
Prefer R? See this tutorial, which provides a similar solution but uses R, and can eb run in either SQL Server 2016 or SQL Server 2017.
The process of building an end to end solution typically consists of obtaining and cleaning data, data exploration and feature engineering, model training and tuning, and finally deployment of the model in production. Development and testing of the actual code is best performed using a dedicated development environment, such as these Python tools:
- PyCharm, a popular IDE
- Spyder, which is included with Visual Studio 2017 if you install the Data Science workload
- Python Extensions for Visual Studio.
After you have created and tested the solution in the IDE, you can deploy the Python code to SQL Server using Transact-SQL stored procedures in the familiar environment of Management Studio.
In this walkthrough, we'll assume that you have been given all the Python code needed for the solution, and you'll focus on building and deploying the solution using SQL Server.
Download the sample dataset and all script files to a local computer.
Execute a PowerShell script that creates a database and a table on the specified instance, and loads the sample data to the table.
Perform basic data exploration and visualization, by calling Python from Transact-SQL stored procedures.
Create new data features using custom SQL functions.
Build and save the machine learning model, using Python in stored procedures.
After the model has been saved to the database, call the model for prediction using Transact-SQL.
We recommend that you do not use SQL Server Management Studio to write or test Python code. If the code that you embed in a stored procedure has any problems, the information that is returned from the stored procedure is usually inadequate to understand the cause of the error.
This walkthrough uses the well-known NYC Taxi data set. To make this walkthrough quick and easy, the data is sampled. Using this data, you'll create a binary classification model that predicts whether a particular trip is likely to get a tip or not, based on columns such as the time of day, distance, and pick-up location.
This walkthrough is intended for users who are already familiar with fundamental database operations, such as creating databases and tables, importing data into tables, and creating SQL queries.
All Python code is provided. An experienced SQL programmer should be able to complete this walkthrough by using Transact-SQL in SQL Server Management Studio or by running the provided PowerShell scripts.
Before starting the walkthrough, you must complete these preparations:
- Install an instance of SQL Server 2017 with Machine Learning Services and Python enabled (requires CTP 2.0 or later).
- The login that you use for this walkthrough must have permissions to create databases and other objects, to upload data, select data, and run stored procedures.