v0.0.9
- Optimize build process and handle concurrent build of the same repo: this will fix issues related to experiment getting stack in the build process and sometimes preventing the iterative hyperparameters algorithms from continuing the search.
- Add advanced nodes scheduling for experiments and jobs: Previously users could set a default node selectors for core and experiments. Now they have the ability to override these selectors per experiment, experiment job for distributed runs, notebook, and tensorboard.
- Fix upload command on Windows platform
- Add more edge cases handling for resumed experiments
- deleting original should delete resumed experiments
- resuming a resumed experiment should resume the original experiment
- Fix issue with stopping experiment groups
- Add NFS provisioner, this is practical for testing the platform without the need to create a real NFS server.
- Add event management and backbone for creating notifications and integrations
- Add tracker
- Add activity log
- Update docs requirement
v0.0.8
- Add restart/resume/copy experiments
- Fix issue with cyclic redirection when the user’s session is expired.
- Add platform logging
- Update Bayesian optimization
- Fix prospector for tests
- Remove requirement for project name in specifications
- Add google GKE tutorial
v0.0.7
- This new version has breaking changes:
- Update specification
- Add kind section to indicate the type of operation to run on Polyaxon.
- Rename sequential search to grid search
- Update early stopping schema to use optimization instead of higher: maximize | minimize
- Move matrix to settings
- Add config for search algorithms to settings
- Remove run_type and export strategies form settings
- All early_stop to stop both individual experiment from running or an experiment group or both.
- Rename steps to build_steps in run section
- Rename concurrent_experiment to concurrency
- Add new method for sampling values
- Possibilty to provide a seed
- Rename Sequential search to grid search
- Add new hyperparameters search algorithms:
- Add hyperband
- Add bayesian optimization [alpha]
- Add continuous methods for sampling matrix params:
- pvalues
- uniform
- quniform
- loguniform
- qloguniform
- normal
- qnormal
- lognormal
- qlognormal
- Add first version of pipelines.
- Show logs after experiment is finished [api, ui, client, cli].
- Optimize plugin jobs [tensorboards and notebooks].
- Separate dahsboard and runner.
- Fixed bug issues.
v0.0.6
- Enhance single experiment spawner
- Add MXNet spawner and support for distributed MXNet experiments
- Add Pytorch spawner and support for distributed Pytorch experiments
- Add Horovod spawner and support for distributed Horovod experiments [WIP]
- Optimize Tensorflow spawner and distributed Tensorflow experiments
- Optimize schedulers’ logic
- Optimize docker build process
v0.0.5
- Add early stopping and number of experiments
- Use statuses for jupyter notebooks and tensorboards
- Fix issue with starting multiple tensorboards
- Rename deleted status to stopped
- Add admin dashboard
v0.0.4
- Add Jupyter notebooks
- Update Tensorboard deployments
- Apply project permission to project plugin jobs, i.e. tensorboard and notebooks will only be accessible to users with enough project permissions
- Update dashboard and ui
- Fix some issues with Kubernetes resources tracking