rfcx / arbimon-jobs

Apache License 2.0
0 stars 0 forks source link

About branches - clarification #13

Open rafael-alvarez opened 9 years ago

rafael-alvarez commented 9 years ago

master Updated 27 days ago --one commit ahead of master-3modeltypes stable Updated 2 hours ago -- THIS BRANCH WILL BE THE STABLE ONE hmm-framed-features Updated 4 days ago --experimental master-old Updated 17 days ago --a version that eliminates binary threshold and multi-rate support master-3modeltypes Updated a month ago --snapshot with slow fast and search and match ransacs Updated 2 months ago --snapshot when ransac (search match) algorithm was finished ssim-plus-k-means Updated 2 months ago --snapshot of SSIM(slow) and Fast algorithms roi_catz Updated 2 months ago --experimental new-soundscape-features Updated 3 months ago

rafael-alvarez commented 9 years ago

Currently the branch that is used in production is master-old.

The master-old branch was taken from before many of the structural changes were implemented. But the master-old branch has the algorithm version (simply put) that works.

However the master-old branch does not have many of the bug fixes nor does not have the modularization implemented. It still uses python pipping to run process in classification. This behaviour was changed in other updated branches.

The intention is to use the stable branch. The stable branch is modulated and has many bug fixes corrected. But need revision in the algorithm department.

g-i-o- commented 9 years ago

Hay varios branches que se pueden cambiar por tags (master-3modeltypes, ransacs, ssim-plus-k-means). Esencialmente la diferencia entre un branch y un tag es que el tag es fijo (como un alias al commit), mientras que el branch permite seguir commiteando.

lazychino commented 9 years ago

@rafael-alvarez @g-i-o- we should do something about this ASAP, because the later the more difficult the merge will be. I think verifying the lastest repo to make sure is production worthy should be a priority.

Also we should have a specification on how to do new models, so creating a new version of an current algorithm don't have to break the previous.(maybe this needs its own issues)