Knowledge Studio 2021.2 Now Available
Knowledge Studio is Altair’s full spectrum AI and machine learning solution. It operationalizes mountains of data to help you make decisions that you can trust.
Knowledge Studio 2021.2 is now available to all licensed users and we recommend that all customer organizations plan their upgrades now. You can find a high-level summary of the improvements in this release attached to this post.
Here’s a high-level summary of the changes in Knowledge Studio 2021.2:
- Substitute Missing Values Node: Performs the substitution of missing values for multiple variables at a time, with an additional option to substitute infinite values +/- {infinity} in each variable with its finite maximum or minimum value. A separate substitution method can be chosen for each variable type. This new capability provides an important alternative to the substitution features in the Variable Transformation node as it substitutes the values in the original variables rather than creating new variables and does this for multiple variables at a time. It also saves time when preparing data for other functions where input data without missing values is preferred.
- Handle Class Imbalance Node: Balances out the classes of a dependent variable with skewed distribution via oversampling the minority class by creating artificial data points (synthesizing data). This node uses Imbalanced-Learn (imblearn) - a popular Python library dealing with classification of imbalanced classes. This feature requires Docker. This new feature substantially expands the choice of sampling methods by adding popular methods including SMOTE, ADASYN, BorderlineSMOTE, KMeansSMOTE, SVMSMOTE, and SMOTENC, combined with a selection of sampling strategies.
- Simpson's Paradox Detection Node: Data profiling feature detects Simpson's Paradox in data (also known as Yule–Simpson effect, amalgamation paradox, or reversal paradox). The paradox manifests itself when a specific trend in the target variable is observed in subsets of data but the opposite trend is observed in the dataset as a whole. Detecting this phenomenon helps avoid the misinterpretation of trends and drawing wrong conclusions about biases or causal relationships between variables. Analysts can make specific adjustments to their predictive models to avoid misleading results.
- Generalized Linear Models: Users can now build models using the new Sklearn GLM node, which uses linear model methods of scikit-learn - a popular Python library for machine learning. The Sklearn GLM Validation node and the Sklearn GLM Scoring node perform validation and scoring for this model. The Sklearn GLM node allows users to build regression models with dependent variables that do not necessarily follow normal distributions. Analysts can fine-tune their models using a better choice of regressors and advanced parameters. (Note: These features require Docker.)
- R Code Generation for Advanced Models: Applies to regression models, regularization, cluster analysis, deep learning, factor analysis, principal component analysis, reject inference, scorecards, and market basket analysis models. In previous versions, R code generation was enabled only for Decision Trees, Strategy Trees, and Ensemble Trees; this version 2021.2 and above, it is enabled for all types of native Knowledge Studio models.
- New Ways to Manage Application Preferences: Users can apply application preferences at the time of installation using silent install. Users can also import or export preferences using the Tools | Preferences menu. This makes it easy to maintain the same application preferences through the upgrade process and saves time for analysts who wish to customize their preferences. Users can keep several preference profiles available for loading in different business contexts and share customized preferences with coworkers.
Find all the download links on the Altair One website.
This link takes you to the Altairone.com marketplace. You must log in to access the release notes, other documentation, and installation files.
Contact us to schedule a meeting or if you have any problems upgrading your implementation.