v1.1.0

v1.1.0

This release adds testing, improves creation and deletion scripts, significantly improves the menu interface, and also clarifies all documentation.

Features

• Added integration tests in Travis (#268)

• Added automated Docker build with Travis (#198).

• Automatically destroy Redis's persistent disks during cluster deletion (#199).

• ReadTheDocs support and full documentation overhaul (#265, #287, #292).

• Improved Docker image pull efficiency (#258).

• Use a randomized bucket/cluster name by default (#212).

• Further generalize helm charts to be overridden with helmfiles (#190).

• Configure default node types to minimize CPU counts (#193).

• Use default-models as the default bucket in tf-serving to provide a set of default models for new clusters (#193).

• Update redis-consumer and frontend to use pre-defined workflows (#179).

• Refactor default queue names from predict and track to segmentation and tracking (#213).

• Added make command for easy Prometheus and associated CRD destruction (#222).

• Added contributor docs to project (#227, #245).

Bugfixes

• Solved issue with inconsistent HPA deployment (#254)

• Reduce errors during menu dialog (#207, #208, #211, #218, #225, #232, #233, #236, #242, #247, #259, #270).

• Prevent infinite loop by only retrying functions 3 times (#205).

• Updated the NVIDIA driver URL (#191).

Breaking Changes

• Removed the QR code for authentication (#196).

Affable Andy

This version of the Deepcell Kiosk was used to obtain the benchmarking results in the preprint Dynamic allocation of computational resources for deep learning-enabled cellular image analysis with Kubernetes.

The Deepcell Kiosk currently supports segmentation of individual live-cell images and tracking of cells across image stacks out of the box,. Furthermore, the architecture is flexible enough that, by extending the kiosk-redis-consumer repository, users can adapt the Deepcell Kiosk to support arbitrary image processing workflows. As example of this extensibility is available in the aforementioned preprint, where we implemented an augmented microscopy workflow.

Valid inputs to the Deepcell Kiosk include individual images or zip files of multiple images.

The Deepcell Kiosk is fully functional on Google Kubernetes Engine and autoscales using Kubernetes' built-in HPA functionality. These features will be supported on Amazon Web Services in a future release.

Add GKE support

Kiosk now works with GKE and AWS.

0.2.1

changed branch from gpus to master (#9)

0.2.0: Add support for GPU instance groups (#6)

* Add support for GPU instance groups

* Install nvidia drivers

* Fix error handling in wait-for-kops. Fix patching so that it overwrites.

* add nvidia-docker-installer.sh script

* Upgrade geodesic. Fix wait-for-kops formatting.

* Upgrade docker-ce for docker-nvidia compatibility

* verbose bash execution

* reduce default size back to zero

0.1.0: Update README.md (#7)

* Update README.md