The currently available version of LityxIQ is 5.1.0, released on 1/24/2021
Functionality updates for 5.1.x:
- LityxIQ now includes a Gurobi-enhanced upgrade, allowing for solving of Optimization problems using a Gurobi solver.
- Updates and improvements to the user-interface, including detailed user-oriented explanation of constraint definitions.
- Scoring Jobs
- Jobs will process batches of records in parallel if allowed by the LityxIQ instance setup.
- New option to select the version/iteration of the model to use for the scoring job. Models in production will still always use the official production version. Note that this means that a model not in production can have multiple scoring jobs each using a different version/iteration.
- New option (on the Advanced tab) to allow a scoring job with no records passed in to process without error. Currently, if no records are in the dataset to be scored, the scoring job throws an error. That will continue to be the default operation, but the user can now specify that it is ok for there to be zero records.
- Machine Learning
- DeepNet algorithm is now fully released.
- Data Manager
- New aggregation functionality includes creating percent of records and percentage of sums, as well as unique value counts, by aggregation group.
- Options are now available to import SAS, SPSS, or Stata datasets.
Functionality updates for 5.0.x:
- The DeepNet neural net algorithm is now available to all users in Beta testing. Please provide feedback on your use of the algorithm in your machine learning models.
- Additional catching and attempted fixing of import errors, and provision of detailed information on the data line causing any import errors.
- Variable names can now contain up to 128 characters (from 100). See https://support.lityxiq.com/378119-Variable-Names for more on variable names in LityxIQ.
- Imported datasets provided with more than 128 characters in a variable name will be automatically reduced to the first 128 characters to become legal.
- Addition of powerful fuzzy matching functions (in addition to existing regular expression pattern matching functions) to Data Manager. See https://support.lityxiq.com/419250-Fuzzy-Matching-Functions for more information.
- An interactive ROC curve and error cost analysis is available for all binary classification models. See https://support.lityxiq.com/364317-C-Statistic-and-ROC-Curves for more information.
- The XGBoost machine learning algorithm is now available to all users.
- Additional performance metrics are computed for both continuous variable models and binary classification models. See https://support.lityxiq.com/289834-Performance-Analysis-Metrics-to-Analyze-Classification-Models and https://support.lityxiq.com/050028-Performance-Analysis-Metrics-to-Analyze-Numeric-Prediction-Models for a complete list of metrics computed by LityxIQ.
- Automated Insights are now available. For any dataset in LityxIQ, simply select up to five variables for which to generate automated insights each time the data is refreshed. Automated Insights include a ranking of which other variables are important predictors of the target, and provide an easy-to-understand multivariate segmentation for each of the selected targets. See https://support.lityxiq.com/528203-What-are-Automated-Insights to get started.
- Variable importance scores and other pre-processing information about machine learning model runs is now available for all algorithms.
- Dataset metadata is now computed in parallel to dataset execution being finalized. This provides much improved execution times for certain large datasets.
- Date-based schedules now allow for hourly jobs.
- The export dataset option allows the selection of an escape character for use in the export file.
- Google BigQuery and Amazon DynamoDB are now available as data sources.