Release v0.2.4

v0.2.4

What's new?

• Introduced a new events view.
• Introduced a new view for viewing network layout, (reducer, combiner and clients hierarchy)
• Introduced a new setup guide-phase to ensure prereqs like package and model are set before starting execution.
• Introduced a better form for parameter selection on run configuration.
• Introduced async dispatching of run configurations.
• Introduced async update refresh of several important fields for user convenincence like status, events, network hierarchy etc.
• Introduced a new download-client-config function to allow for faster and more convenient client configuration.
  (Just download config and point your local client and whoallah! You are online in this federation.)

Other

• Fixed logic bugs related to framework persistance.
• Fixed a logic bug causing clients to get assigned prior to compute package assignment (and hence will not account for assignment policy).
• Fixed a logic bug if reducer is resumed from previous state (to ensure) that the right compute package is selected.
• Update dependency versions.

v0.2.3

What's new?

• Support for latest Minio
• Improvements i UI - now not possible to submit jobs is in monitoring state.
• Improvement of Docker image hierarchy.

Other:

• Docs updates
• Several bugfixes and security patches.

v0.2.2

v0.2.2

What's new?

• The MNIST examples (Keras and PyTorch) have been updated so that they now bundle the example data in .npz format.

Other

• Docs updates

v0.2.1

v0.2.1

What's new?

• It is now possible to choose which validation metrics to plot in the Dashboard

Fixes

• Plots backed by no current data is no longer shown as empty plots.

Other

• Docs updates

v0.2.0

What's new?

• It's now possible to have examples in external repositories
• Support for models constructed with the Keras Functional API
• Set maximum number of clients in the settings file

combiner:
name:
combinerhost:
combinerport: 12080
max_clients: 50

• Added visualizations on FEDn communication performance to the dashboard
• Added client allocation policy to spread the clients evenly over the combiners
• Use config for s3 model commits instead of a hard-coded bucket name
• Memory management to prevent combiners from going off
• Now possible to upload the compute package through the UI
• Reducer, client and combiner now have their own Dockerfile definitions

Fixes

• Combiners now handle the case when all clients fail to update a model

Other

• Lots of product documentation updates

v0.1.4

• Fixed volume path to prevent combiner crashes from writing every temporary model to RAM memory, and hitting RAM max memory limit.

v0.1.3

• Resolves a bug in the calculation of the average in the combiner which would be affecting smaller models and a few client cases

v0.1.2

Additions:

• Added new plot for time/round
• Added CPU loads and MEM plot for all rounds
• Allocate clients to accepting combiner with the least number of clients
• Monitoring CPU/MEM/ROUNDS with personalized plots
• Added HTML documentation templates, now accessible from FEDn

Fixes:

• Removed flask-dashboard dependency which was previously bloating the Docker image
• Combiner now handles the case when all Clients fails to update the model
• Removed usage of hard-coded "models" bucket for s3 model commits, now using config instead

v0.1.1

Added:

• Pull Request template
• Bug Report and Feature Request templates
• Contribution Guide
• Code of Conduct

v0.1.0

Major

Compute package bundling, distribution and execution.

• Ability to toggle remote distribution

Reducer init sequence.

• Ability to re(initialize) reducer from command line and settings.yaml
• Control state_store

Tempfile storage backend

• combiner can now choose a tempfile backend for file storage of interim models.

Minor

Many performance improvements and bug fixes. See complete change log for details.