Niftynet meeting 7th august 2017

Attendance Tom Doel, Wenqi Li, Dzhoshkun Shakir, Tom Whyntie, Robert Gray, Guotai Wang, Carole Sudre, Felix Bragman, Ben, Jorge Cardoso, Eli Gibson, Irme Groothuis

1. Question pip install. Pb with tensorflow-gpu or other requirement when use on laptop when tensorflow gpu does not work.
   1. Solution1: Duplicate package: niftynet-gpu and niftynet / Solution2: pip installer with option (tensorflow gpu and tensorflow cpu) 2. Custom importers
   2. Command script in addition to remove gpu install cpu
   3. Drop tensorflow installation from the pip installer. Similarly to tensorflow: niftynet and niftynet-gpu

ADOPTED: Removing tensorflow from the installation but create warning asking people to install tensorflow if need be. Link to tensorflow installation page. Try catch with import

2. Naming convention:

   1. net_${task}
   2. verb for actions: segment, regress, embed, classify, reconstruct.
   3. individual sub executables if need afterwards

3. Testing framework - PEP-8 pylint convention - Use of Auto PEP-8 package

   1. Solution Merge dev with PEP-8 and rebase
   2. Each of contributor to rebase and run AutoPEP8
   3. Policy: No merging if no pep8

4. Timeframe for release

   1. 21st August Features freeze latest merge
   2. 25th August Inclusion of Refactoring
   3. 1st September Freeze of task specific
   4. Afterwards Cleaning only

5. Continuous integration

   1. Testing pip-install
   2. GitHub Mirror. NiftK account (only public master and personal branches) No dev on GitHub. Master corresponds to last stable version
   3. Dev branch stays public, accessible on CMICLab
   4. Automatic commit tag version_release and specific tags
   5. Use similar tagging system as other nifty releases (cf niftyreg using command --version
   6. Need for test to be run on Windows machine as well
   7. Need to separate tests in two populations (quick ones to be run by everyone on separate branches and the more expensive ones only when merging on dev)

6. Features:

   1. At least segmentation and GAN. Evaluation as part of segmentation

7. Model Zoo:

   1. Idea of a package manager for binaries based on python with version control on the binaries

8. Website

   1. Small typos to be adressed to Tom W.

9. Demos

   1. To be done using Eli's example as guideline

10. Niftynet Journal club

    1. Every 3rd Monday of the month: informal, for idea sharing. 2 papers, 15 minutes each.
    2. First instance 21st August: Felix B, M. Jorge C.