Added a new Feature Extractor Screen that provides an efficient way to prepare, analyze, transformed sensor data, and extract features before training a model
Added AutoML and Custom Training screens
Added a feature statistic to data feature extraction pipeline steps
Minor Features
Added Sample Rate capture setting as part of the metadata project settings
Bug Fixes
Fixed issue where nearly added Augmentation didn’t have the option to add a new step after
Fixed issue with stopping task at the Test Model screen
Fixed capture selection workflow at the Test Model screen
Fixed issue where Pipeline Augmentation steps didn’t survive import/export of pipeline
Fixed issue where clicking on the model in model results does not take to the correct model in the Model Explore screen
The Advanced Settings tab has been moved to the bottom of the action buttons.
Enhanced Model Building Status Check to retry when a network issue occurs instead of returning a failure message.
Bug Fixes
Fixed an issue with the Query Filter in the Prepare Data view. If one of the metadata values is 0, the value was being removed from the query filter.
Fixed the Default Query Selection in the Build Model view for projects with a non-custom segmenter.
Fixed an issue where the magnitude column was duplicated when using Query Segments and Custom Feature Generator set.
Feature Threshold selection was not being persisted on the pipeline. This issue has been fixed.
The rename feature in the Knowledge Pack tab in the Project Summary view, renames the selected model’s name. In the case of a hierarchical model, all child model names needed to be prefixed with the parent model name. This issue has been fixed.
Fixed an issue with the Feature Vector Distribution Graph in the Explore Model View. The chart was not changing to the child model’s data when users switched to the child model tabs.
For hierarchical models, the parent model KP description details were not being posted back to the server with the Test Model run request. This issue has been fixed.
Added Pipeline Summary to the Explore Model Screen, showing the pipeline flow and step details.
Added Knowledge Pack Summary to the Explore Model Screen, showing the steps and step details used to generate the Knowledge Pack.
Added variance and correlation based feature thresholds to the advanced settings in the Model Building view.
Enhanced the Projects table by adding a segment count column. Starter Licenses are limited to 2500 segments per project. For Starter Licenses, the segments column will show a usage indicator of the number of segments used for the project.
Moved all summary tables into a single tab view and defaulted to displaying 10 rows per table.
Create and Modified Date columns are added to the Queries and Pipelines Summary tables.
Added a Knowledge Pack counts column to the Pipeline Summary table showing the number of Knowledge Packs for the pipeline.
Knowledge Pack summary table is enhanced to show Accuracy, Classifier, Model Size and Feature Counts.
Moved the “Refresh Summary” button to the bottom of the tab view.
Changed default sorting on the Query and Pipeline Summary tables to sort by the latest last modified date column and the Knowledge Pack summary table by the created date column.
Enhanced the Captures Files Table to show metadata data marked for lookup as drop-down filters.
Applied color gradient based on the ratio of value to support in the Confusion Matrix.
Title shows the Project Name on all screens except for the Home Screen.
Bug Fixes
Fixed issue in the Download Model view where the default Target OS selection for Cortex M4 was not correct.
Enhanced Project and Capture Statistics tables with advanced filtering capabilities and type aware filter options (Date, Text, and Numeric Filters).
Added Feature Cascade, Strip Mean, and Magnitude Transforms to the Model Building View.
Added ability to select a Knowledge Pack Architecture in the Model Building View. You can select between a Single or Hierarchical Multi-Model or both.
Added an enhanced experience for Model Testing.
Test Model menu displays a new section for model testing.
You can now sort and filter, capture files on all capture attributes and metadata fields.
Multiple capture files can be selected to simultaneously run signal recognition.
Added a menu for selecting/unselecting all capture files.
Enabled selection of any session and model on the project to run the capture files to run signal recognition and compute the accuracy.
Accuracy is displayed in the table after signal recognition.
Capture row is color highlighted to show if accuracy is above 80% (green) or below 80% (red) or grey (when no ground truth data is provided).
Added ability to compute a summary confusion report of selected capture files run’s.
Enhanced Model Test Results.
Classification Results are displayed when the results button is clicked.
The results section consists of a confusion matrix, a classification chart showing the ground truth and predictions for the segments, and a feature vector heat map showing the feature values for a given feature and segment.
Enabled synchronous hover on the classification chart and feature vector heatmap, when hovered over the heat map the predicted event for the segment is displayed in the classification chart.
Enhanced Confusion Matrix.
Diagonal cells in the Matrix with data are highlight green and non-diagonal data cells are highlighted red.
UNC column has been removed from the table and Sense Perc will only be displayed when Support is greater than 0.
Accuracy will only be displayed when there is ground truth data.
All decimal values in the matrix are rounded to 2 decimals.
