Support for more complicated checkpoints logic

• Renamed axes argument to axis throughout the library
• Significantly faster sampling sources in dpipe.batch_iter
• More extensions are supported by load and save from dpipe.io

Process-based parallelism in batch iterators

v0.0.9

v0.0.9

A collection of small updates combined into a release

• Renamed dpipe.medim to dpipe.im. Moved all medical-related code to separate repositories. dpipe.medim will still be importable until the next release
• Support for adaptive batch size in batch iterators. To ensure reproducibility, sources now have a random_state argument
• Experiment layouts - a new, simpler way to build and run experiments
• Various fixes

alpha-6

• Got rid of Snakemake.
• Added functions with bindings to the configs grammar.
• Added dicom_utils.
• Removed a lot of deprecated stuff and reorganized some modules' structure.

alpha-5

• Removed tensorflow
• Added more sugar in the configs' grammar
• Refactored some preprocessing code
• Moving towards modular models and batch-predict

alpha-4

• Major refactoring of datasets
• Improved configs grammar
• Snakemake files are modular now
• Added setup.py. Now dpipe can be installed via pip

alpha-3

• Simplified the batch iterators
• Updated external resources support
• Improved configs grammar

alpha-2

highly cutomized the training process. updated the docs.
made the resource manager even more pythonic.
added support for external resources.
moved documentation htmls to another repo.

alpha-1

Merge pull request #32 from neuro-ml/develop

Develop