Release Notes for Machine Learning - Application

February 9, 2023

Version 2023.02

Manage Columns

You can now easily select the columns you want to include in the modeling process and change column data types. To use this feature, select Manage Columns in the Prep Data stage.

January 31, 2023

Version 2023.01

Data and Feature Anomaly Checks Added to Export and Predict

Added a message to inform you if your data for prediction contains categorical values that were not present in your training data . The message details what those values are, and which columns contain them.

Customizable Feature Selection

Added the ability to select which features you use in your model. This includes engineered features. To customize your feature selection, go to the Features tab in the Feature Engineering step . Reducing your feature count can speed up the modeling process.

Partial Dependence Plot Performance Improvements

Improved the speed of populating the Partial Dependence plot. On average, you can expect 50% faster load times.

December 8, 2022

Version 2022.41

Runtime Versioning

Added the ability to upgrade individual projects to the latest AYX ML Runtime Version—and to revert back to your most recent version if you change your mind. You can find this functionality in the 3-dot menu on each project on the project page.

• Project versioning ensures that changes to modeling algorithms DO NOT impact business results unless you select an upgrade. This provides a consistent experience when you revisit a previous project.
• Note that the AYX ML Runtime Version upgrade only applies to model tuning improvements made by Alteryx. Improvements to the UX, new features, security, and infrastructure update automatically.

December 1, 2022

Version 2022.40

Time Series Enhancements

What Is Time Series 

Time Series regression expands your modeling capabilities with data that includes a time component. Now, you can forecast into the future and get accurate predictions. Do things like demand forecasting, financial forecasting, and more with Alteryx Machine Learning. 

Alteryx Machine Learning uses commonly used and state-of-the-art time series models. These include Facebook Prophet, ARIMA, and ETS, in addition to other regression models such as XGBoost and LightGBM.  

Why Time Series Matters 

Use Time Series for future-looking, time-based predictions. This empowers you to quickly leverage prior data to forecast future outcomes. Enhanced capabilities account for trending and seasonality, making model performance stronger.  

New Functionality  

Time Series now includes 2 major enhancements to the Machine Learning experience: 

1. Decomposition [Model Setup step]: Visualize trend and seasonal signals​ in isolation from the residual signal. This allows the non-time series specific models to perform better in most cases​ (where time-series specific models are Facebook Prophet, ARIMA, and ETS). We run all models with and without decomposition and then display the best model on the leaderboard. We support Decomposition Visualizations for these frequencies: 

• Hourly 

• Daily 

• Weekly  

• Monthly 

• Quarterly 

1. Time Series Forecast Graph and Data Export [Export and Predict step]: Introduced line graph and graph data to visualize and then use forecasted data.  

Changes to Existing Workflow  

For a clearer workflow, we moved items from Data Insights to a new Model Setup step. Note that this doesn’t represent new functionality, just a new organizational flow. 

August 25, 2022

Version 2022.28

Feature Engineering Step 

Added an automated Feature Engineering step to Alteryx Machine Learning. This step allows you to apply primitives (generalized operations) to calculate new features. These features can improve model performance and help you gain further insight. Additionally, you can view the correlation values for the engineered features. Look for this new feature on the left side of the Alteryx Machine Learning interface.

August 2, 2022

Version 2022.25

Updated Integer Parsing

We extended support for parsing real numbers (for example, 2.45 and 3.0) as integers. The parsing truncates the right side of the decimal point. For example, 2.45 becomes 2, and 3.0 becomes 3.

New Time Series Primitives

Added "Fourier Transform" and "Rolling Trend" primitives for improved Time Series feature engineering. We also added an additional 16 DateTime primitives for feature engineering.

Custom Input for DateTime Formats

Added 20 new DateTime formats and the ability to enter your own custom format. You can now use more DateTime datasets with our Time Series Regression model.

Machine Learning Predict Tool—Data Validation for Time Series Models

Added Time Series data validation to the Machine Learning Predict tool in Designer. The data validation checks if your input data length is longer than the forecast horizon.

Correlation Tab—Switch X-Y Axes on 2-Variable Plot

Added an option to switch the axes of the 2-Variable Plot in the Correlations tab.

July 21, 2022

Version 2022.24

Pipeline Highlights 

Expanded Pipeline Highlights in the Auto Model leaderboard for categorical and numerical operations. We now sort operations by order of execution in the model pipeline. Use this information to choose a model based on which operations we use in the model.

June 23, 2022

Version 2022.20

Time Series—Irregularly Spaced Data

Support added for irregularly spaced Time Series datasets.

Time Series—Data Checks

Added data checks to warn you when we modify your data.

June 16, 2022

Version 2022.19

Stop Criteria—Model Setup

You now have the option to set "time" as the stop criteria for model search completion. The minimum stop time is 1 minute.

April 18, 2022

Version 2022.12

Facebook® Prophet Estimator

Added Prophet to improve predictions for Time Series Regression models. This addition targets datasets with trends and seasonality.

ARIMA Regressor

Added ARIMA Regressor for Time Series Regression. ARIMA is a high performant statistical algorithm for Time Series datasets.

Select Completed Models

Updated the Stop button in the Auto Model search step. You can now stop the model search and select completed models up to that point.

Features Visibility

Updated the Features panel in the Auto Model step to show all features contributing to your model. This helps you understand and evaluate models based on the features in use. In addition, features used in ensemble models are also shown.

March 17, 2022

Version 2022.08

Unary Column Detection

A new data check identifies unary data type columns and automatically recommends next steps. This enables you to quickly detect columns to exclude from Machine Learning. The No Variance Data Check flags columns that contain only 1 unique value and then provides you with a recommendation to drop those columns.

February 2, 2022

Version 2022.03

Alteryx Machine Learning is now available!