Release 1.14
- 16 Nov 2023
- 1 Minute to read
- Print
- DarkLight
- PDF
Release 1.14
- Updated on 16 Nov 2023
- 1 Minute to read
- Print
- DarkLight
- PDF
Article summary
Did you find this summary helpful?
Thank you for your feedback
Python SDK release - October 30, 2023
Python SDK for API access to Data Explorer has been released, and details are present here.
Release 1.14.96 (Maintenance release) - October 27, 2023
- Bug fix for an issue where the search was failing for certain values of tunables used as search space reduction criteria.
Release 1.14.92 - October 16, 2023
- Enhanced clustering controls and snapshot: With this release, the clustering capabilities have become richer than ever, with the ability to store the clustering state and retrieve it later. Additionally, the clusters can be named. For example, during your exploration of an animal dataset, if you find that a 'green' colored cluster mostly has cats in it, then you can name the cluster 'cats', and this name can be persisted and retrieved every time you visualize the job. Please see Clustering controls for more details.
- Enhanced 'Highlight points' view: The 'Highlight points' view is enhanced with points colored based on their cluster membership and controls provided to modify the group by criteria. In the screenshot below, the 'Highlight points' on search results show the spread of results in the feature space and the clusters that these results belong to.
- Support resultset upload for 'Explore-Sequence' jobs: A resultset created in an Explore-Sequence type of job contains a video sequence for each point added to the resultset. This release supports uploading these sequences as a video or as individual frames using 'adectl resultset upload' command. Please see Resultsets for more details on uploading resultsets.
- New featurizers
- dino-v2 based ingestion pipelines for images and videos are available. These pipelines support a maximum job size of 35K sequences for Explore-Sequence jobs and 100K images/frames for all other jobs.
- Akridata[Image/Video]ClipPipelineV2 introduces an enhanced CLIP model that gives better quality results on text search.
- License note: The pipeline to dataset attachment operation requires you to read and accept the licenses for pipelines that are based on externally sourced featurizers.
Additional notes
Resubmit old visualizing jobs
With this release, any visualization job that was created before 1.10.nnn release will go into a DECAYED state, as shown below.
To visualize this job, please click on 'RESUBMIT' button marked above.
To visualize this job, please click on 'RESUBMIT' button marked above.
Was this article helpful?