What's Changed

• minor fixes by @SheldonGrant in #109
• Update README.md by @SheldonGrant in #110
• dataset task logging and predict not require response file by @SheldonGrant in #111
• move assert magic to separate utilty by @SheldonGrant in #114

Full Changelog: V0.9.1...v0.9.2

What's Changed

• add streamlined example templates by @SheldonGrant in #107

Full Changelog: v0.9.0...V0.9.1

• flexible routines for build, train, deploy
• mldock stages for stage management
• setting routines per stage (i.e. dev, prod, etc)
• build routines default to mldock local build behaviour
• train and deploy routines are automatically at project init, however they only function for --interactive mode. This can be great for running differently in development.

What's Changed

• brings routines as flexible commands to mldock by @SheldonGrant in #92
• bugfix on models/datasets create error when creating new by @SheldonGrant in #92
• clean up private methods from utils.py to be normal public methods by @SheldonGrant in #92
• roll version to v0.8.25 by @SheldonGrant in #93

Full Changelog: v0.8.24...v0.8.25

What's Changed

• Feature/advanced predict functionality by @SheldonGrant in #84
• change container => project command group by @SheldonGrant in #87
• Bug/datasets add compression zip by @SheldonGrant in #89

Full Changelog: v0.8.23...v0.8.24

• adds basic dataset and models management
• adds central management of remotes in mldock/config

Better container init

• check available templates in template server
• defaults github templates to mldock template server on github, if configured for github. No need for API key.
• resolves initialize a container from provided trainer, prediction and requirements.txt files #66 - init container from provided train, prediction and requirements files

Better development

• providing an --interactive on train/deploy which runs those commands without docker

streamlines base containers:

• migrating to a fastapi + uvicorn server
• stripping out non-base requirements (mldock logger, csv handling in request) - current containers are a cleaner basis to start from.
• splitting src/container/assets.py => src/container/lifecycle.py + src/assets.py. To help with explainability and "migratability" meaning that user altered code is not removed in a container init --container-only execution.
• brings a @Wrap decorator to provide user scripts with appropriate wrappers for error handling and startup/cleanup tasks in container lifecycle.

bugfixes:

• requirements.txt handling in container init where src/requirements.txt already existed would fail due to create=True on packagemanager.

• package pack is a wrapper around pip wheel in hopes to streamline dependency management in container projects
• bugfix: logs metrics/params groks updated to be more flexible on metric naming. Specifically to include special characters by using a GREEDYDATA groking approach.