Release 0.15

[mfeurer]

No description provided.

[codecov]

Codecov Report

Merging #1561 (013d7ee) into master (b2ac331) will decrease coverage by 3.09%.
The diff coverage is 82.69%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1561      +/-   ##
==========================================
- Coverage   88.05%   84.96%   -3.10%     
==========================================
  Files         140      155      +15     
  Lines       10995    11898     +903     
  Branches        0     2058    +2058     
==========================================
+ Hits         9682    10109     +427     
+ Misses       1313     1244      -69     
- Partials        0      545     +545     

[eddiebergman]

So turns out the major issue and why there was so many merge conflicts is beacuse I created a seperate branch when creating v0.14.4 which was not included in development. Therefore master and development disagreed on that state of things as this v0.14.4 commit was not in development, despite having all the same content except release notes.

My strategy to fix everything was

# master: -----* ---A --B --C --M
# devlop:       \-------------------D
# * = v0.14.7
# A, B, C useful commits from master


# Just stick development on top of the HEAD of master but all conflicts are `dev` favoured
git checkout development
git rebase -Xtheirs master

# This leaves us in a valid state to merge with master as it includes it but
# we've discarded any useful commits that caused a merged conflict

# master: -----* ---A --B --C --M
# devlop: -----*'---A'--B'--C'--M'----------D
# *', A', B', C' = the corrupted commits from master

# We then replay each commit from `v0.14.4` on master all the way up to its `HEAD` to replay
# the commits one by one, while being in a valid merge state on develop 

# `v0.14.4` is the biggest culprit of merge conflicts (700+),
# I simply took the release notes from here and favoured dev for everything else,
# as it should have only included content from dev at that time + new release notes.
git cherry-pick -m 1 v0.14.4

# master: -----* ---A --B --C --M
# devlop: -----*'---A'--B'--C'--M'----------D-* 

# For the remaining commits, like A', B', C', I replayed them on top of the branch D so 
# I can replay the changes and make sure it's all included
git cherry-pick -x -n A'..M'

# master: -----* ---A --B --C --M
# devlop: -----*'---A'--B'--C'--M'----------D-*--A--B--C--M

# We now have develop and master in a mergable state and the important
# changes after `v0.14.4` where things diverged are now incorporated propely.

[eddiebergman]

Few points:

• Doc failed to build, I think I missed a link that I need to update and I will do that
• The meta-features failing test i"m not sure the root cause but it seems it's trying to index into a SingleBestModel which is the dummy model.
• There seems to be an issue with the latest branch according to one user. callback function error! #1569 (comment). The issue isn't about callbacks but rather some meta-learning and data dtypes.

[eddiebergman]

Actually, it's nothing to do with a model but it seems in the publish docker build workflow specifically, it seems to have some incorrect code? It seems to me like here info is set to DummyDataManager

if info["task"] in  REGRESSION_TASKS:

TypeError: 'DummyDatamanager' object is not subscriptable

E       CompletedProcess(args='python3 /workspace/scripts/02_retrieve_metadata.py --working-
directory /tmp/autosklearn-unittest-tmp-dir-99ac6cd59ac2-7412-46734 ', returncode=1, 
stdout=b'binary.classification accuracy 0\n', stderr=b'Traceback (most recent call last):\n  
File "/workspace/scripts/02_retrieve_metadata.py", line 274, in <module>\n    main()\n  File 
"/workspace/scripts/02_retrieve_metadata.py", line 237, in main\n    configuration_space = 
pipeline.get_configuration_space(\n  File "/usr/local/lib/python3.8/dist-packages/autosklearn
/util/pipeline.py", line 47, in get_configuration_space\n    if info["task"] in 
REGRESSION_TASKS:\nTypeError: \'DummyDatamanager\' object is not subscriptable\n')