Hydra is a framework for elegantly configuring complex applications

Facebook Research

Last update: Jan 4, 2023

Related tags

Configuration hydra

Overview

A framework for elegantly configuring complex applications.

Check the website for more information.

Releases

Stable

Hydra 1.0 is the stable version of Hydra.

Documentation
Installation : pip install hydra-core --upgrade

Development release

Hydra 1.1 is still in active development and should not be considered stable. You are welcome to try it out at your own risk and report any issues.

Documentation
Installation : pip install hydra-core --upgrade --pre
Dev releases are more likely to have bugs, please report any issues.
APIs and behaviors are subject to change without notice between different dev releases.

License

Hydra is licensed under MIT License.

Community

Ask questions in the chat or StackOverflow (Use the tag #fb-hydra or #omegaconf):

Citing Hydra

If you use Hydra in your research please use the following BibTeX entry:

@Misc{Yadan2019Hydra,
  author =       {Omry Yadan},
  title =        {Hydra - A framework for elegantly configuring complex applications},
  howpublished = {Github},
  year =         {2019},
  url =          {https://github.com/facebookresearch/hydra}
}

Comments

Use circleCI API to kick off multiple pipelines for different plugins.
update 09/17 CircleCI support helped me find a way to kick off pipelines for forked PRs (after a few back and forth emails.), yayy.

Test plan:

on forked PR

This PR is a forked PR. Pls check the checks sections for all the jobs that was kicked off. one example here, notice the plugin test was kicked off by beloved mecha-hydra.

IN Hydra repo

I created a PR from within Hydra, the checks can be seen in this PR. https://github.com/facebookresearch/hydra/pull/985

update 09/16

[x] move github token to github secrets https://github.com/facebookresearch/hydra/issues/973

[x] use deploy key instead of user key in hydra I've removed the user key and created a read only deploy key for Hydra cirleCI and kicked off a build here https://app.circleci.com/pipelines/github/facebookresearch/hydra/3484/workflows/d822ee36-a0bb-4a04-9ccd-73a42c2b4db8 (this build runs all the workflows, so if it succeeds it means we are good.)

[x] some related config needed for forked PR.

Then we should be good to go with this PR.

update 09/01

We found a good solution for running tests in forked PR. I'm going to

[x] address omry's comments.

[x] reach out to oss with our plan - done, task in progress

[x] submit feature request to circleCI on only sharing selected secrets - done, support ticket created.

TODO

[x] Create github bot user and set it up in Hydra (blocked by github support, our bot account was flagged as spam.)

Motivation

Try out parameterized the nox session.

reduced the CI time from 46m to 16m (disclaimer: one datapoint only)

Have you read the Contributing Guidelines on pull requests?

Yes/No

Test Plan

validate yaml locally, and CI

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)
CLA Signed
opened by jieru-hu 40

Ray launcher WIP

Update 10/14 Follow up items:

[x] add custom resolver to get lib versions.
[x] clean up orphaned instances.

Update 09/30

A few items I want to follow up separately (will create PR for these)

cannot run the plugin in python 3.6 (cloudpickle issue, same error message as #428, failed run here https://app.circleci.com/pipelines/github/jieru-hu/hydra/611/workflows/0d99e3b0-1442-446f-857b-f476a7707b6d/jobs/6648 ), I tried a few things (updating cloudpickle version etc but was not able to resolve it.)
nightly builds test AMIs (right now the AMI is set as an env variable which is annoying everytime when we need to update the AMI id, I want to automate this.)
doc update. The doc needs to be refreshed a bit. I think it might be easier to create a separate PR for this.

Update 09/28

Summary of the changes:

1. Address omry's comments.

2. Changes to integration test:

The goal is "No outbound traffic for the test instances." The barrier is the pip install and conda create we need to run while setting up the instance which requires us to open 443 to all outbound traffic. To get around this: for conda and dependency packages needed for starting the cluster, I created a base AMI that has everything pre-installed. 2) for Hydra related packages, build the wheels at test time and install the wheels on the instance. The upside is we achieve "no outbound traffic for the instance", the downside is that means we need update AMI when dependencies changes. To help with that I created a script (create_ami.py) to automate building the AMI.

It would be good to build nightly AMIs and wheels, that's something I want to work on soon.

output from running `create_ami.py`

$ AWS_PROFILE=jieru python create_ami.py 
2020-09-28 16:23:56.268051 - Running: aws ec2 authorize-security-group-egress --group-id sg-0a1 --ip-permissions IpProtocol=tcp,FromPort=443,ToPort=443,IpRanges=[{CidrIp=0.0.0.0/0}]
2020-09-28 16:23:57.464861 - 
2020-09-28 16:23:57.487688 - Running: ray up /var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/tmpjw6lihef.yaml -y
2020-09-28 16:25:57.400029 - 2020-09-28 16:23:58,540    INFO cli_logger.py:388 -- Using cached config at /var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/ray-config-d951a214f8602b878335411b5df6e84af463922b

2020-09-28 16:25:57.462567 - Running: ray rsync_up /var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/tmpjw6lihef.yaml './setup_ami.py' '/home/ubuntu/' 
2020-09-28 16:25:59.210039 - 2020-09-28 16:25:58,320    INFO cli_logger.py:388 -- Using cached config at /var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/ray-config-d951a214f8602b878335411b5df6e84af463922b
2020-09-28 16:25:58,736 INFO cli_logger.py:388 -- NodeUpdater: i-0d82abc901a725abc: Syncing ./setup_ami.py to /home/ubuntu/...
2020-09-28 16:25:58,922 INFO log_timer.py:25 -- NodeUpdater: i-0d82abc901a725abc: Got IP  [LogTimer=186ms]
building file list ... done
setup_ami.py

sent 692 bytes  received 42 bytes  489.33 bytes/sec
total size is 1345  speedup is 1.83
Installing dependencies now, this may take a while...
2020-09-28 16:25:59.210121 - Running: ray exec /var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/tmpjw6lihef.yaml 'python ./setup_ami.py' 
...
Installing collected packages: typing-extensions, omegaconf
Successfully installed omegaconf-2.0.2 typing-extensions-3.7.4.3
2020-09-28 23:45:09.147903 - OUT: /home/ubuntu/anaconda3/envs/hydra_3.8.5/bin/pip install antlr4-python3-runtime==4.8
2020-09-28 23:45:09.927774 - OUT: Processing ./.cache/pip/wheels/c8/d0/ab/d43c02eaddc5b9004db86950802442ad9a26f279c619e28da0/antlr4_python3_runtime-4.8-py3-none-any.whl
Installing collected packages: antlr4-python3-runtime
Successfully installed antlr4-python3-runtime-4.8
2020-09-28 23:45:09.927847 - OUT: /home/ubuntu/anaconda3/envs/hydra_3.8.5/bin/pip install --ignore-installed PyYAML
2020-09-28 23:45:10.798501 - OUT: Processing ./.cache/pip/wheels/13/90/db/290ab3a34f2ef0b5a0f89235dc2d40fea83e77de84ed2dc05c/PyYAML-5.3.1-cp38-cp38-linux_x86_64.whl
Installing collected packages: PyYAML
Successfully installed PyYAML-5.3.1
Shared connection to 34.221.119.106 closed.
2020-09-28 16:45:11.007294 - Running: aws ec2 revoke-security-group-egress --group-id sg-0a1 --ip-permissions IpProtocol=tcp,FromPort=443,ToPort=443,IpRanges=[{CidrIp=0.0.0.0/0}]
2020-09-28 16:45:12.047395 - 
2020-09-28 16:45:12.047395 - 
ec2.Image(id='ami-0c46') current state pending
ec2.Image(id='ami-0c46') current state pending
...
ami-0c46 ready for use now.

#### 3. skip `-Werror` flag for ray launcher, the tests will fail with the flag, stack trace (this is caused by ray, not the plugin itself) - solution is to add a pytest.ini in ray's tests dir to suppress the warnings.

Stack trace

test_ray_local_launcher.py .[2020-09-28 21:52:46,553][HYDRA] Ray Launcher is launching 1 jobs, sweep output dir: /private/var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/pytest-of-jieru/pytest-25/test_sweep_1_job_ray_local_ove0
[2020-09-28 21:52:46,553][HYDRA] Initializing ray with config: {'num_cpus': 1, 'num_gpus': 0}
2020-09-28 21:52:46,564 INFO resource_spec.py:223 -- Starting Ray with 8.79 GiB memory available for workers and up to 4.41 GiB for objects. You can adjust these settings with ray.init(memory=<bytes>, object_store_memory=<bytes>).
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis-shard_0.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis-shard_0.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis-shard_0.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis-shard_0.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/redis.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/gcs_server.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 748, in start_head_processes
    self.start_gcs_server()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/gcs_server.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/gcs_server.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 748, in start_head_processes
    self.start_gcs_server()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/gcs_server.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/monitor.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 750, in start_head_processes
    self.start_monitor()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/monitor.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/monitor.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 750, in start_head_processes
    self.start_monitor()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-46_557592_90415/logs/monitor.err' mode='a' encoding='utf-8'>
F[2020-09-28 21:52:47,565][HYDRA] Ray Launcher is launching 2 jobs, sweep output dir: /private/var/folders/n_/9qzct77j68j6n9lh0lw3vjqcn96zxl/T/pytest-of-jieru/pytest-25/test_sweep_2_jobs_ray_local_ov0
[2020-09-28 21:52:47,565][HYDRA] Initializing ray with config: {'num_cpus': 1, 'num_gpus': 0}
2020-09-28 21:52:47,573 INFO resource_spec.py:223 -- Starting Ray with 8.74 GiB memory available for workers and up to 4.38 GiB for objects. You can adjust these settings with ray.init(memory=<bytes>, object_store_memory=<bytes>).
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis-shard_0.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis-shard_0.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis-shard_0.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis-shard_0.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 746, in start_head_processes
    self.start_redis()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/redis.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/gcs_server.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 748, in start_head_processes
    self.start_gcs_server()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/gcs_server.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/gcs_server.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 748, in start_head_processes
    self.start_gcs_server()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/gcs_server.err' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/monitor.out' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 750, in start_head_processes
    self.start_monitor()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/monitor.out' mode='a' encoding='utf-8'>
Exception ignored in: <_io.FileIO name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/monitor.err' mode='ab' closefd=True>
Traceback (most recent call last):
  File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/node.py", line 750, in start_head_processes
    self.start_monitor()
ResourceWarning: unclosed file <_io.TextIOWrapper name='/tmp/ray/session_2020-09-28_21-52-47_566018_90415/logs/monitor.err' mode='a' encoding='utf-8'>

INTERNALERROR> Traceback (most recent call last):
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/main.py", line 191, in wrap_session
INTERNALERROR>     session.exitstatus = doit(config, session) or 0
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/main.py", line 247, in _main
INTERNALERROR>     config.hook.pytest_runtestloop(session=session)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/hooks.py", line 286, in __call__
INTERNALERROR>     return self._hookexec(self, self.get_hookimpls(), kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 93, in _hookexec
INTERNALERROR>     return self._inner_hookexec(hook, methods, kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 84, in <lambda>
INTERNALERROR>     self._inner_hookexec = lambda hook, methods, kwargs: hook.multicall(
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 208, in _multicall
INTERNALERROR>     return outcome.get_result()
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 80, in get_result
INTERNALERROR>     raise ex[1].with_traceback(ex[2])
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 187, in _multicall
INTERNALERROR>     res = hook_impl.function(*args)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/main.py", line 272, in pytest_runtestloop
INTERNALERROR>     item.config.hook.pytest_runtest_protocol(item=item, nextitem=nextitem)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/hooks.py", line 286, in __call__
INTERNALERROR>     return self._hookexec(self, self.get_hookimpls(), kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 93, in _hookexec
INTERNALERROR>     return self._inner_hookexec(hook, methods, kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 84, in <lambda>
INTERNALERROR>     self._inner_hookexec = lambda hook, methods, kwargs: hook.multicall(
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 208, in _multicall
INTERNALERROR>     return outcome.get_result()
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 80, in get_result
INTERNALERROR>     raise ex[1].with_traceback(ex[2])
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 187, in _multicall
INTERNALERROR>     res = hook_impl.function(*args)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/runner.py", line 85, in pytest_runtest_protocol
INTERNALERROR>     runtestprotocol(item, nextitem=nextitem)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/runner.py", line 100, in runtestprotocol
INTERNALERROR>     reports.append(call_and_report(item, "call", log))
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/runner.py", line 188, in call_and_report
INTERNALERROR>     report = hook.pytest_runtest_makereport(item=item, call=call)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/hooks.py", line 286, in __call__
INTERNALERROR>     return self._hookexec(self, self.get_hookimpls(), kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 93, in _hookexec
INTERNALERROR>     return self._inner_hookexec(hook, methods, kwargs)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/manager.py", line 84, in <lambda>
INTERNALERROR>     self._inner_hookexec = lambda hook, methods, kwargs: hook.multicall(
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 203, in _multicall
INTERNALERROR>     gen.send(outcome)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/skipping.py", line 129, in pytest_runtest_makereport
INTERNALERROR>     rep = outcome.get_result()
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 80, in get_result
INTERNALERROR>     raise ex[1].with_traceback(ex[2])
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/pluggy/callers.py", line 187, in _multicall
INTERNALERROR>     res = hook_impl.function(*args)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/runner.py", line 260, in pytest_runtest_makereport
INTERNALERROR>     return TestReport.from_item_and_call(item, call)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/reports.py", line 294, in from_item_and_call
INTERNALERROR>     longrepr = item.repr_failure(excinfo)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/python.py", line 1511, in repr_failure
INTERNALERROR>     return self._repr_failure_py(excinfo, style=style)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/nodes.py", line 355, in _repr_failure_py
INTERNALERROR>     return excinfo.getrepr(
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 635, in getrepr
INTERNALERROR>     return fmt.repr_excinfo(self)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 880, in repr_excinfo
INTERNALERROR>     reprtraceback = self.repr_traceback(excinfo_)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 824, in repr_traceback
INTERNALERROR>     reprentry = self.repr_traceback_entry(entry, einfo)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 774, in repr_traceback_entry
INTERNALERROR>     source = self._getentrysource(entry)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 685, in _getentrysource
INTERNALERROR>     source = entry.getsource(self.astcache)
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/code.py", line 246, in getsource
INTERNALERROR>     astnode, _, end = getstatementrange_ast(
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/_pytest/_code/source.py", line 384, in getstatementrange_ast
INTERNALERROR>     astnode = ast.parse(content, "source", "exec")
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/ast.py", line 47, in parse
INTERNALERROR>     return compile(source, filename, mode, flags,
INTERNALERROR>   File "/Users/jieru/opt/anaconda3/envs/pytest38/lib/python3.8/site-packages/ray/worker.py", line 869, in sigterm_handler
INTERNALERROR>     sys.exit(signum)
INTERNALERROR> SystemExit: 15
mainloop: caught unexpected SystemExit!

4. install conda in circleCI linux docker

Previously we pin the circleCI linux docker image to be python:3.8. however, the image runs on python 3.8.6 which is not yet available to be installed in conda. As a result, the ray launcher tests fails (cloudpickle requires the exact same version of python used on pickle and unpickle side) To be consistent with how tests are run in MACOS and WIN, I added the miniconda installation for linux machines as well. The installation takes a few secs, so I didn't add cache for it.

Update 09/21

Now that #815 has finally landed. This PR is unblocked finally! This is my priority this week. TODO items.

[x] rebase onto latest master
[x] upload latest wheels to S3 for installation during integration tests. In order to upload and install latest wheels in the integration, I want to:

list all the plugins that's going to be tested (by getting the PLUGINS env variable), build wheels, and scp them all to the ec2 instance.
Install all the wheels on the ec2 instance. This way, we can remove all outbound traffic of the testing ec2 instances.

[x] Address omry's comments
[ ] Update circleCI test user creds. and finish all the TODO items outlined in the proposal quip.

Update 09/08 edit: moved the TODO items to the latest update.

Update 09/01

This PR has been blocked by #815. Now that we've figured out a good solution for #815, I will go ahead and get 815 in first and then circle back here.

Also I'm going to create data class for both ray init, ray remote and boto configs (we will only add typing for common boto fields, and the boto config will extend Dict[str, Any])

For the integration tests to work on the latest code, we are going to build wheels and upload to S3 with each integration test run. This will be likely be a a separate PR.

Motivation

This is built on https://github.com/facebookresearch/hydra/pull/515, sorry I had to open a new pull requests. I still need to figure out what's a better workflow with forking & syncing.

This PR address some comments from PR515:

Add local mode for ray launcher
Add docker options
refactor the Launcher class, group all file syncing together.

Plan to add in next PR(s):

Add Integration test
Add an option for users to update the cluster if they need.
Remote cluster return results to laptop.
make _dump_func_params better/less hard coded.

Update 04/10 Add Integration tests Ray up automatically update cluster so no need for us to provide an option for update. Now JobReturns are copied back to laptop refactor _dump_func_params a bit more.

Fix LOCAL mode to run RAY directly.
Add integration tests for both LOCAL and AWS mode.

Have you read the Contributing Guidelines on pull requests?

Yes/No

Test Plan

Integration tests Run launcher in both local and AWS mode

Run the

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)

CLA Signed

opened by jieru-hu 38

[Bug] I can't see any log with pytorch DDP

🐛 Bug

Description

When I use PyTorch's Distributed Data Parallel I can't see any logs in system out and log file is empty except wandb log like below.

[2020-11-06 01:32:08,629][wandb.internal.internal][INFO] - Internal process exited

Checklist

[x] I checked on the latest version of Hydra
[x] I created a minimal repro

To reproduce

You can reproduce by running this repo(https://github.com/ryul99/pytorch-project-template/tree/a80f0284c5b22fba2d4892bb906a9bc2b6075838) with python trainer.py train.working_dir=$(pwd) train.train.dist.gpus=1(DDP with one gpu) I'm sorry that this is not that minimal code 😅 ** Stack trace/error message **

❯ python trainer.py train.working_dir=$(pwd) train.train.dist.gpus=1
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/hydra/core/utils.py:204: UserWarning: 
Using config_path to specify the config name is deprecated, specify the config name via config_name
See https://hydra.cc/docs/next/upgrades/0.11_to_1.0/config_path_changes
  warnings.warn(category=UserWarning, message=msg)
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/hydra/plugins/config_source.py:190: UserWarning: 
Missing @package directive train/default.yaml in file:///home/ryul99/Workspace/pytorch-project-template/config.
See https://hydra.cc/docs/next/upgrades/0.11_to_1.0/adding_a_package_directive
  warnings.warn(message=msg, category=UserWarning)
Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to dataset/meta/MNIST/raw/train-images-idx3-ubyte.gz
100.1%Extracting dataset/meta/MNIST/raw/train-images-idx3-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw/train-labels-idx1-ubyte.gz
113.5%Extracting dataset/meta/MNIST/raw/train-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to dataset/meta/MNIST/raw/t10k-images-idx3-ubyte.gz
100.4%Extracting dataset/meta/MNIST/raw/t10k-images-idx3-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw/t10k-labels-idx1-ubyte.gz
180.4%Extracting dataset/meta/MNIST/raw/t10k-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw
Processing...
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/torchvision/datasets/mnist.py:480: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:141.)
  return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)
Done!

Expected Behavior

If you run with python trainer.py train.working_dir=$(pwd) train.train.dist.gpus=0(not using DDP), you can see many logs like this.

❯ python trainer.py train.working_dir=$(pwd) train.train.dist.gpus=0
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/hydra/core/utils.py:204: UserWarning: 
Using config_path to specify the config name is deprecated, specify the config name via config_name
See https://hydra.cc/docs/next/upgrades/0.11_to_1.0/config_path_changes
  warnings.warn(category=UserWarning, message=msg)
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/hydra/plugins/config_source.py:190: UserWarning: 
Missing @package directive train/default.yaml in file:///home/ryul99/Workspace/pytorch-project-template/config.
See https://hydra.cc/docs/next/upgrades/0.11_to_1.0/adding_a_package_directive
  warnings.warn(message=msg, category=UserWarning)
[2020-11-06 01:48:33,990][trainer.py][INFO] - Config:
train:
  name: First_training
  working_dir: /home/ryul99/Workspace/pytorch-project-template
  data:
    train_dir: dataset/meta/train
    test_dir: dataset/meta/test
    file_format: '*.file_extension'
    use_background_generator: true
    divide_dataset_per_gpu: true
  train:
    random_seed: 3750
    num_epoch: 10000
    num_workers: 4
    batch_size: 64
    optimizer:
      mode: adam
      adam:
        lr: 0.001
        betas:
        - 0.9
        - 0.999
    dist:
      master_addr: localhost
      master_port: '12355'
      mode: nccl
      gpus: 0
      timeout: 30
  test:
    num_workers: 4
    batch_size: 64
  model:
    device: cuda
  log:
    use_tensorboard: true
    use_wandb: false
    wandb_init_conf:
      name: ${train.name}
      entity: null
      project: null
    summary_interval: 1
    chkpt_interval: 10
    chkpt_dir: chkpt
  load:
    wandb_load_path: null
    network_chkpt_path: null
    strict_load: false
    resume_state_path: null

[2020-11-06 01:48:33,991][trainer.py][INFO] - Set up train process
[2020-11-06 01:48:33,991][trainer.py][INFO] - BackgroundGenerator is turned off when Distributed running is on
Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to dataset/meta/MNIST/raw/train-images-idx3-ubyte.gz
100.1%Extracting dataset/meta/MNIST/raw/train-images-idx3-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw/train-labels-idx1-ubyte.gz
113.5%Extracting dataset/meta/MNIST/raw/train-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to dataset/meta/MNIST/raw/t10k-images-idx3-ubyte.gz
100.4%Extracting dataset/meta/MNIST/raw/t10k-images-idx3-ubyte.gz to dataset/meta/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw/t10k-labels-idx1-ubyte.gz
180.4%Extracting dataset/meta/MNIST/raw/t10k-labels-idx1-ubyte.gz to dataset/meta/MNIST/raw
Processing...
/home/ryul99/.pyenv/versions/LWD/lib/python3.8/site-packages/torchvision/datasets/mnist.py:480: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:141.)
  return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)
Done!
[2020-11-06 01:48:38,886][trainer.py][INFO] - Making train dataloader...
[2020-11-06 01:48:38,905][trainer.py][INFO] - Making test dataloader...
[2020-11-06 01:48:40,366][trainer.py][INFO] - Starting new training run.
[2020-11-06 01:48:40,467][train_model.py][INFO] - Train Loss 2.3010 at step 1
[2020-11-06 01:48:40,473][train_model.py][INFO] - Train Loss 2.3133 at step 2

System information

Hydra Version : 1.0.3
Python version : 3.8.5
Virtual environment type and version : pyenv 1.2.21
Operating system : Ubuntu server 18.04

Additional context

Add any other context about the problem here.

bug question

opened by ryul99 34

[Feature Request] [Submitit-Plugin] (Potentially a bug) Impossible to set certain flags in submitit launcher
🚀 Feature Request

Motivation

This could be read as a feature request or a bug report, I'm not sure how you want to consider it, I'm going with feature request. Currently theres some flags that you cannot set in the submitit launcher. I practice "gpu centric" scheduling, so I like to specify mem_per_gpu and cpus_per_gpu and then I can just use gpus_per_task to always get the optimal settings.

For example on "mystery cluster" (you know the one), we can use 10 cpus and 64GB RAM per GPU. These settings allow me to only have to change gpus_per_task for example if I need to have 2 GPUs for 2 different models, then I'll automatically get 20 cpus and 128GB RAM for each of the tasks without having to change all of the settings. I've already PRed stuff related to this into submitit.

The problem occurs when you try to set, via additional parameters, something like mem_per_gpu. You can't set mem and mem_per_gpu, slurm just crashes when you do that. Similarly, if you try to set cpus_per_gpu via additional_parameters you'll wind up setting that in addition to cpus_per_task.

Pitch

I tried making a simple patch that fixes this, but it hits issues with the AutoExecutor which I never updated in submitit to be aware of the options which conflict with each other. I think in general auto is missing some of the recent work in submitit, and it feels like a semi-hacky workaround to me.

Submitit already has pretty good validation logic so my pitch is to (1) let submitit handle everything by calling the correct executor for the job instead of using Auto. This is going to require a revamp of how the parameters are named/passed however and will likely be a breaking change to the API. (2) It would be nice if we could allow people to pass whatever parameters are supported by submitit without needing to update the hydra schema each time. These are my two major goals, so they sound reasonable/feasible? Point (2) may not be possible I guess.

Other options:

Use auto but try to include our own validation logic (needs exploration), I don't like this because it requires us to duplicate logic that submitit has already and also it may not even be possible depending on what Auto decides to do (looked iffy on my quick pass)

Punt this to submitit to revamp the auto executor to do the right stuff and be more flexible. I don't like this because I think the auto executor is more of a convenience hack and more complex/flexible code would be calling the executors directly (this is what I was doing previously with my own submitit hydra wrapper thing). I also think its going to be a more more complex PR.

Are you willing to open a pull request? (See CONTRIBUTING)

Yes, but I want feedback first on the best way to go about it

Additional context

Add any other context or screenshots about the feature request here.
enhancement
opened by Queuecumber 33
upgrade isort to v5.5.2
Motivation

Fix issue #751

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

(How should this PR be tested? Do you require special setup to run the test or repro the fixed bug?) Run tests as per the contributing documentation.

$ nox -s lint $ FIX=1 nox -s lint_plugins $ pytest $ pre-commit run --all-files # to check the new isort hooks pass

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax) issue #751
CLA Signed
opened by vporta 33
Add a Nevergrad Sweeper plugin
Motivation

This is a proof of concept for a Nevergrad sweeper pluging.

I won't have time to work much on it so if anyone is willing to take it from there, feel free :D The current concerns I have:

the documentation should be improved, in particular how to choose an optimizer. The simplest solution is "OnePlusOne" for now. It's incredibly simple and very robust, so I made it the default optimizer.

I don't have ideas for a nice commandline syntax. There are a lot of possible ways to define the parametrization of a problem in nevergrad (probably too many, then again, it's a research framework so we need some flexibility for now :s). I've included a hack to be able to use the full nevergrad syntax, but that won't be very robust :D . For now only choices and standard bounded scalars are supported without the hack, we would at least need a way to define log-distributed scalars.

the launchers seem to only support batches, I suppose there is no way to check if some evaluations have finished so that we could make use of the free workers and launch more evaluations without waiting? That could make the sweep faster, and as far as I know with close to no performance loss.

Minimizing on integers is probably not as robust as minimizing on floats, I'm a bit concerned that anyone writting 0:10 in the commandline would get integers only and may not realize it, maybe the default should be float, and we'd need another way to cast to integers.

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

Unit tests are included, as well as a runnable example

Related Issues and PRs

Closes #330
CLA Signed
opened by jrapin 33
Add Joblib Launcher for parallel execution (multi-process and multi-threaded)

Motivation

Launcher for local parallel sweeps using Joblib.Parallel, as proposed in #357, and discussed with @omry.

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

Added tests following the example_launcher_plugin.

Related Issues and PRs

Implements #357
CLA Signed

opened by jan-matthis 32
Add Ax plugin for Hydra

Motivation

Hydra Plugin to interface with Adaptive Experimentation Platform

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

To be added

Related Issues and PRs

https://github.com/facebookresearch/hydra/pull/291
CLA Signed

opened by shagunsodhani 32
[Ax Sweeper]: Integrate override parser
Motivation

Integrate override parser.

Todo:

[x] Clarify how to integrate fixed-params

[x] Update documentation

[x] Remove type inference logic

[x] Add news entry

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

CI

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)

#870
CLA Signed
opened by shagunsodhani 30
Allow configuring the search path via the config file

Design doc outlining consideration and possible implementations:

https://docs.google.com/document/d/1Dx77SThg1ugnGHvZ8Dq1FQusXDvGn0tyyWBeJ3nXvro/edit?usp=sharing
enhancement

opened by omry 30
Fix for docker support in Ray AWS Launcher

Motivation

This PR fixes this issue , and patches a small bug that crashes Docker containers launched with the plugin. Some redundant hidden behavior has been removed from EnvSetupConf.

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

All the core tests pass, and the same number of plugin tests fail as on the main branch (22 passed, 20 errors). Testing containerized jobs managed with Ray on EC2 is pretty tricky, I'm not sure how to design a production-level test for something like that. You'd need an AWS account and something like DockerHub serving an image Ray could download.

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)

https://github.com/facebookresearch/hydra/issues/1191
CLA Signed

opened by samuelstanton 26
[Bug] MissingConfigException cannot be correctly deserialized, due to lack of missing_cfg_file ctor default
🐛 Bug

Description

in https://github.com/facebookresearch/hydra/blob/main/hydra/errors.py the missing_cfg_file parameter of the MissingConfigException should be defaulted to None since it is optional, otherwise deserialization will fail.

Checklist

[x] I checked on latest commit [7bc2b1a] of errors.py (https://github.com/facebookresearch/hydra/commit/7bc2b1ad66da91a12c6158f9413c908b211bff1e)

[x] I created a minimal repro (See this for tips).

To reproduce

** Minimal Code/Config snippet to reproduce ** import hydra e=hydra.errors.MissingConfigException('missing', 'file') import pickle x=pickle.dumps(e) y=pickle.loads(x) ** Stack trace/error message **

Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: __init__() missing 1 required positional argument: 'missing_cfg_file'

Expected Behavior

successful deserialization:

>>> y MissingConfigException('missing')

System information

Hydra Version : hydra-core==1.3.1

Python version : Python 3.8.13

Virtual environment type and version : None

Operating system : Ubuntu 22.04.1 LT

Additional context

This exception was serialized/deserialized when using ray tune.
bug
opened by gazonk 0
[Documentation Bug] Use of the term 'duck type' in documentation

This may be me being a stickler but I was super confused by the use of the term duck-typing on this (and related pages):

https://hydra.cc/docs/tutorials/structured_config/minimal_example/

The official Python documentation refers to this functionality as type hinting. I understand duck typing as a concept is all about testing for functionality rather than being explicit about type. So in short, I'm super confused about the documentation using the term 'duck typing' when referring to the cfg: MySQLConfig code.

That being said the official Python page also uses the term duck typing in various places on the page, but the overarching term to me still appears to be type hinting, unless I'm missing something.
bug

opened by Gazoo101 0
Remove usages of legacy importlib API.

Motivation

The Python >= 3.10 compatible usages of importlib added in 277ff54b71e31531923a51459b328391d76bb5b3 can be safely applied to any version of Python >= 3.5 which covers the full range that Hydra supports.

Getting rid of the old deprecated API, asides from cleaning up, will keep Hydra compatible with PyInstaller if PyInstaller doesn't get cold feet and revert its dropping of support for the legacy API (https://github.com/pyinstaller/pyinstaller/pull/7344).

Have you read the Contributing Guidelines on pull requests?

Yes

Test Plan

(How should this PR be tested? Do you require special setup to run the test or repro the fixed bug?)

Standard CI/CD testing should be sufficient.

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)

None.
CLA Signed

opened by bwoodsend 1
docs: reorganize sections in Configuring Hydra page

This PR reorders sections on the https://hydra.cc/docs/1.3/configure_hydra/intro/ webpage. I've moved the section on top-level hydra config settings higher up on the page, and I've merged the section on hydra.mode into the section on top-level hydra settings. Additionally, I've added a clarifying sentence to the entry regarding. hydra.mode.
CLA Signed

opened by Jasha10 0
$improve hydra.utils.get_{class,method,object}$
improve hydra.utils.get_{class,method,object}
Implement hydra.utils.get_object for an easier way to look up objects from a dotpath.

Commits:

improve err msg for get_class failure

improve tests for get_{class,method}

implement hydra.utils.get_object (Closes #2139)

document hydra.utils.get_{class,method,object} (Closes #1972)

CLA Signed
opened by Jasha10 0

[Bug] instantiate: top-level `_target_` has inconsistent behavior with `_convert_`

🐛 Bug

Description

The _convert_ flag has no effect when instantiating a config that has a _target_ at the top level.

To reproduce

from dataclasses import dataclass

import hydra
from omegaconf import DictConfig
from pytest import raises

@dataclass
class MyDataclass:
    ...

cfg = {"_target_": "__main__.MyDataclass"}

obj = hydra.utils.instantiate(cfg, _convert_="none")
with raises(AssertionError):
    assert isinstance(obj, DictConfig)  # got MyDataclass

obj = hydra.utils.instantiate(cfg, _convert_="partial")
with raises(AssertionError):
    assert isinstance(obj, DictConfig)  # got MyDataclass

obj = hydra.utils.instantiate(cfg, _convert_="object")
assert isinstance(obj, MyDataclass)  # Ok

obj = hydra.utils.instantiate(cfg, _convert_="all")
with raises(AssertionError):
    assert isinstance(obj, dict)  # got MyDataclass

Expected Behavior

None of the above assertions should raise an AssertionError; _convert_ should have the same effect for top-level configs as for nested configs.

System information

Hydra Version : 1.4.0.dev0
Python version : 3.10.8

Additional context

Based on bug report #2504.

bug instantiate_api structured config

opened by Jasha10 0

Releases(v1.3.1)

v1.3.1(Dec 21, 2022)
1.3.1 (2022-12-20)

This bugfix release updates a version pin on the OmegaConf library, allowing Hydra to be installed alongside the latest version of OmegaConf.

Bug Fixes

Relax OmegaConf pin allowing OmegaConf 2.3 to be installed (#2510)

Links:

Hydra on pypi

Hydra 1.3 docs

Source code(tar.gz)
Source code(zip)
v1.3.0(Dec 8, 2022)
1.3.0 (2022-12-08)

Features:

Implement _convert_="object" option for instantiate, enabling conversion of non-_target_ structured configs to instances of the backing dataclass / attr class. (#1719)

Enable layering of the @hydra.main decorator on top of other decorators produced using functools.wraps. (#2303)

Allow for non-leading dashes in override keys (#2363)

support specifying an absolute path with --config-path (#2368)

Support python3.11 (#2443)

Bug Fixes:

Fix an issue where Hydra's exception-handling logic could raise an AssertionError (#2342)

Links:

Hydra on pypi

Hydra 1.3 docs

Source code(tar.gz)
Source code(zip)
v1.2.0(May 17, 2022)
1.2.0 (2022-05-17)

Bug fixes

hydra.runtime.choices is now updated correctly during multi-run (#1882)

hydra.verbose=True now works with multirun. (#1897)

Fix a resolution error occurring when a nested class is passed as a _target_ keyword argument to instantiate (#1914)

It is now possible to pass other callable objects (besides functions) to hydra.main. (#2042)

New features

Add support to Hydra's instantiation API for creation of functools.partial instances via a _partial_ keyword. (#1283)

Support defining basic sweeping in input config. (#1376)

Improve error message with more context when an omegaconf exception occurs during the config merge step. (#1697)

Add --experimental-rerun command-line option to reproduce pickled single runs (#1805)

Add experimental Callback for pickling job info. (#2092)

Implement tab completions for appending to the defaults list (+group=option) and deleting from the defaults list (~group). (#1841)

Enable the use of the pipe symbol | in unquoted strings when parsing command-line overrides. (#1850)

Support for Python 3.10 (#1856)

Improve clarity of error messages when hydra.utils.instantiate encounters a _target_ that cannot be located (#1863)

The instantiate API now accepts ListConfig/list-type config as top-level input. (#1950)

Improve error messages raised in case of instantiation failure. (#2099)

Add callback for logging JobReturn. (#2100)

Support disable changing working directory at runtime. (#910)

Support setting hydra.mode through config. (#394)

Behavior changes

The antlr version requirement is updated from 4.8 to 4.9, to align better with current antlr versions

If user code raises an exception when called by instantiate, raise an InstantiateError exception instead of an instance of the same exception class that was raised by the user code. (#1911)

Remove support for deprecated arg config_loader to Plugin.setup, and update signature of run_job to require hydra_context. (#1953)

The remaining changes are protected by the new version_base support, which allows one to either configure Hydra to support older setups / config, or configure Hydra to use the following more modern defaults:

Remove deprecated "old optional" defaults list syntax (#1952)

Remove support for the legacy hydra override syntax (see deprecation notice). (#2056)

Remove support for old hydra.experimental.{compose,initialize} interface

Remove support for _name_ and _group_ from package header (see deprecation notice)

Remove support for legacy default list interpolation format (see deprecation notice)

Remove support for TargetConf class

Remove support for strict flag from compose API (see deprecation notice)

Remove support for ".yml" extensions, requiring ".yaml" instead.

Default to not changing the working directory at runtime. Use hydra.job.chdir=True to reinstate old behavior.

Default to not adding any directory to the config path. (see config_path options)

Source code(tar.gz)
Source code(zip)
v1.1.2(Apr 12, 2022)
1.1.2 (2022-04-12)

Features

Add support to Hydra's instantiation API for creation of functools.partial instances via a _partial_ keyword. (#1283)

The instantiate API now accepts ListConfig/list-type config as top-level input. (#1950)

Improve error messages raised in case of instantiation failure. (#2099)

Add callback for logging JobReturn. (#2100)

API Change (Renames, deprecations and removals)

Fix an internal key collision to prevent an exception when - group@_global_: choice is used in the same defaults list as - group: choice. (#1784)

Bug Fixes

Improve error message when a missing default must be specified at the command line (#1742)

Fix an edge case where a defaults list defined in a missing dataclass node causes a ConfigCompositionException. (#1765)

Source code(tar.gz)
Source code(zip)
v1.1.1(Aug 19, 2021)
1.1.1 (2021-08-19)

Features

Allow range() in override grammar to have only one argument (the stop value), e.g. range(3) (#1664)

Add support for a HYDRA_DEPRECATION_WARNINGS_AS_ERRORS envorinment variable. When it is set, using deprecated features causes an exception instead of a warning. (#1689)

To make migration from Hydra 1.0 to 1.1 easier, Hydra will now issue a warning if the primary config defines config values and Defaults List if the Defaults List does not specify _self_ (#1755)

API Change (Renames, deprecations and removals)

Re-introduced a deprecated strict flag in the Compose API (#1694)

Relax handling of legacy hydra overrides in the defaults list to make migration from Hydra 1.0 to 1.1 easier (#1748)

Bug Fixes

Allow Defaults List interpolation in nested configs and document that the interpolation keys must be absolute (#1668)

instantiate API now supports target as a parameter name (#1669)

Fix --cfg=job --resolve and --help --resolve so that the ${hydra:...} resolver now prints properly (#1681)

Fix unexpected changes to defaults list in some cases when appending an experiment (+experiment=test) (#1706)

Fixed composition where top level of a config composed from a config group is a list (#1724)

Fix bug where overriding hydra config groups more than once without using the override keyword resulted in an error instead of deprecation warnings (#1730)

Fix failure when sanitizing stack traces resulting from job exceptions (#1739)

Source code(tar.gz)
Source code(zip)
v1.0.7(Jul 15, 2021)
1.0.7 (2021-07-14)

This release contains a few small changes to make Hydra 1.0 a bit more compatible with OmegaConf 2.1 and Hydra 1.1 to enable easier migration. There is no risk in using it if you are using 1.0.6 now as it just makes the migration a bit smoother in some scenarios:

If you are upgrading OmegaConf to 2.1 before upgrading Hydra to 1.1.

It enables some configs to be compatible with both versions in some rare cases related to the change in the default composition order in Hydra 1.1.

Maintenance Changes

Enables Hydra 1.0 to work with OmegaConf 2.1 (Manual installation of OmegaConf 2.1 is required) (#1634)

Hydra 1.0 ignores _self_ in Defaults List for partial future compatibility with Hydra 1.1 (#1712)

Source code(tar.gz)
Source code(zip)
v1.1.0(Jun 10, 2021)
1.1.0 (2021-06-09)

This is the biggest Hydra release yet.

Highlights

OmegaConf 2.1 supports relative interpolations, nested interpolations, more powerful resolvers and better compatibility with plain Python dict and list.

Recursive defaults list: Every config can now have a Defaults List.

Recursive instantiation: When instantiating objects using the instantiate API, nested defined objects are instantiated automatically.

There have also been many bug fixes and performance improvements, as well of as some breaking changes - both in Hydra and in OmegaConf. Please check the API changes and deprecation sections in these release notes and in the OmegaConf release notes.

Features

General enhancements

Upgrade to OmegaConf 2.1. Please check the release notes for it (#1426)

Support for Python 3.9. (#1062)

Improve performance of config composition in a benchmark by 64% (#1328)

Allow @, $ and ? symbols in unquoted values in overrides (#1074, #1437, #1597)

Add a --resolve flag that can be used with the --cfg and --help flags to resolve interpolations before the config is printed. (#1585,#1482)

It is now possible to disable Hydra's logging configuration (#1130)

Support for Zsh tab completion (#347)

Support for force-add of config values via ++key=value (force-add overrides if the value exists and adds it otherwise) (#1049)

New experimental Callback API (#1402)

Config composition enhancements

Support for Defaults List in any config (Recursive relative defaults) (#1170)

Composition order of a config with a Defaults List can be specified with the _self_ keyword. (#326)

Support for configuring the config search path from the primary config (#274)

Add support for selecting multiple configs from the same Config Group (#499)

Final choices of defaults list are retained in the dictionary hydra.runtime.choices (#956)

Object Instantiation enhancements

Support for recursive instantiation with hydra.utils.instantiate() (#566)

Instantiate now fully supports positional arguments (#1432)

Support for converting parameters to primitive containers during instantiation (#1015)

It is now possible to override the _target_ of instantiation with a Python type or a string (#1017)

API Change (Renames, deprecations and removals)

General changes

Default composition order change, see this for details (#1217)

Remove deprecated strict flag from @hydra.main and the Compose API (#1010)

Passing a config name as config_path to @hydra.main is now an error (#1010)

Promote hydra.experimental.{compose,initialize,initialize_config_dir,initialize_config_module} out from the hydra.experimental module to the hydra module. (#1030)

The override grammar now requires that, in quoted strings, any sequence of \ preceding a quote (either an escaped quote, or the closing quote) must be escaped (#1600)

Support partial failures in MULTIRUN mode (#1377)

Add HydraContext to run_job and Launcher/Sweeper's setup (#1498)

Instantiate changes

Instantiate is now recursive by default, use _recursive_=False to disable recursive instantiation. (#566)

Deprecated TargetConf, you should no longer be extending or annotating with it (#1010)

Remove ObjectConf and deprecated params support that was deprecated in Hydra 1.0 (#1010)

Plugins API changes

HydraContext required in run_job, Launcher and Sweeper's setup methods, see issue for details (#1498)

ConfigSourcePlugins needs to be modified to support recursive defaults, see link for details (#1080)

Bug Fixes

Hydra no longer erroneously changes the USER environment variable in pytest unit tests once installed (#1059)

Fix a bug where Structured Config has a field like list : Optional[List[int]] = None (#1117)

No longer modifies exception stack trace when running under a debugger (#1237)

hydra.job.{id,num} are now properly passed to jobs in multirun (#1270)

Add support for %f directive (microseconds) to the ${now:PATTERN} resolver (#1287)

Fixed a bug where tab completion did not work if the Defaults List had a missing (???) item. (#1381)

Fix + overrides to properly insert new values into Structured Configs. (#1515)

Fix edge cases where using the command line to set a key to a value containing a string ending with a backslash could crash (#1600)

--cfg=hydra now works for read-only config (#1461)

Change hydra.job_logging and hydra.hydra_logging to be non-optional (#1656)

Configs with unicode characters are now working on Windows (#1659)

Plugins

Add Optuna Sweeper plugin

Improved Documentation

New Defaults List page (#1170)

New Extending Configs pattern (#1170)

Major updates to the Packages page (#1170)

New Configuring Experiments pattern (#1170)

Documentation updates for configuring search path (#1449)

Documentation for hydra.callbacks (#1526)

Document how to configure Hydra plugins (#1582)

Source code(tar.gz)
Source code(zip)
v1.1.0.rc1(May 13, 2021)
1.1.0.rc1 (2021-05-13)

This is the biggest Hydra release yet.

Highlights

OmegaConf 2.1 supports relative interpolations, nested interpolations, more powerful resolvers and better compatibility with plain Python dict and list.

Recursive defaults list: Every config can now have a Defaults List.

Recursive instantiation: When instantiating objects using the instantiate API, nested defined objects are instantiated automatically.

There have also been many bug fixes and performance improvements, as well of as some breaking changes - both in Hydra and in OmegaConf. Please check the API changes and deprecation sections in these release notes and in the OmegaConf release notes.

Features

General enhancements

Upgrade to OmegaConf 2.1. Please check the release notes for it (#1426)

Support for Python 3.9. (#1062)

Improve performance of config composition in a benchmark by 64% (#1328)

Allow @, $ and ? symbols in unquoted values in overrides (#1074, #1437, #1597)

Add a --resolve flag that can be used with the --cfg flag to resolve interpolations before the config is printed. (#1585)

It is now possible to disable Hydra's logging configuration (#1130)

Support for Zsh tab completion (#347)

Support for force-add of config values via ++key=value (force-add overrides if the value exists and adds it otherwise) (#1049)

Config composition enhancements

Support for Defaults List in any config (Recursive relative defaults) (#1170)

Composition order of a config with a Defaults List can be specified with the _self_ keyword. (#326)

Support for configuring the config search path from the primary config (#274)

Add support for selecting multiple configs from the same Config Group (#499)

Final choices of defaults list are retained in the dictionary hydra.runtime.choices (#956)

Object Instantiation enhancements

Support for recursive instantiation with hydra.utils.instantiate() (#566)

Instantiate now fully supports positional arguments (#1432)

Support for converting parameters to primitive containers during instantiation (#1015)

It is now possible to override the _target_ of instantiation with a Python type or a string (#1017)

API Change (Renames, deprecations and removals)

General changes

Default composition order change, see this for details (#1217)

Remove deprecated strict flag from @hydra.main and the Compose API (#1010)

Passing a config name as config_path to @hydra.main is now an error (#1010)

Promote hydra.experimental.{compose,initialize,initialize_config_dir,initialize_config_module} out from the hydra.experimental module to the hydra module. (#1030)

The override grammar now requires that, in quoted strings, any sequence of \ preceding a quote (either an escaped quote, or the closing quote) must be escaped (#1600)

Instantiate changes

Instantiate is now recursive by default, use _recursive_=False to disable recursive instantiation. (#566)

Deprecated TargetConf, you should no longer be extending or annotating with it (#1010)

Remove ObjectConf and deprecated params support that was deprecated in Hydra 1.0 (#1010)

Plugins API changes

HydraContext required in run_job, Launcher and Sweeper's setup methods, see issue for details (#1498)

ConfigSourcePlugins needs to be modified to support recursive defaults, see link for details (#1080)

Bug Fixes

Hydra no longer erroneously changes the USER environment variable in pytest unit tests once installed (#1059)

Fix a bug where Structured Config has a field like list : Optional[List[int]] = None (#1117)

No longer modifies exception stack trace when running under a debugger (#1237)

hydra.job.{id,num} are now properly passed to jobs in multirun (#1270)

Add support for %f directive (microseconds) to the ${now:PATTERN} resolver (#1287)

Fixed a bug where tab completion did not work if the Defaults List had a missing (???) item. (#1381)

Fix + overrides to properly insert new values into Structured Configs. (#1515)

Fix edge cases where using the command line to set a key to a value containing a string ending with a backslash could crash (#1600)

Plugins

Add Optuna Sweeper plugin

Improved Documentation

New Defaults List page (#1170)

New Extending Configs pattern (#1170)

Major updates to the Packages page (#1170)

New Configuring Experiments pattern (#1170)

Source code(tar.gz)
Source code(zip)
v1.1.0.dev6(Apr 28, 2021)
The following are informal release notes for Hydra 1.1 as of the publication of 1.1.0.dev6. Note that there are many changes in OmegaConf 2.1 that are not covered by these release notes. Once OmegaConf 2.1 is officially release it will have formal release notes.

Features

Support for converting parameters to primitive containers during instantiation (#1015)

It is now possible to override the target of instantiation with a type or a string (#1017)

Support for force-add of config values via ++key=value (force-add overrides if the value exists and adds it otherwise) (#1049)

Support Python 3.9. (#1062)

It is now possible to disable Hydra's logging configuration (#1130)

Support for Defaults List in any config (Recursive relative defaults) (#1170)

Improve performance of config composition in a benchmark by 64% (#1328)

Can now add keys to fields annotated as Dict[K,V] from the command line without + prefix (#1421)

Upgrade to OmegaConf 2.1 (#1426)

Instantiate now fully supports positional arguments (#1432)

Support for configuring the config search path from the primary config (#274)

Composition order of a config with a Defaults List can be specified with the self keyword. (#326)

Support for Zsh tab completion (#347)

Add support for selecting multiple configs from the same Config Group (#499)

Support for recursive instantiation with hydra.utils.instantiate() (#566)

Final choices of defaults list are retained in the dictionary hydra.runtime.choices (#956)

API Change (Renames, deprecations and removals)

Deprecated TargetConf, you should no longer be extending or annotating with it (#1010)

Remove ObjectConf and deprecated params support that was deprecated in Hydra 1.0 (#1010)

Remove deprecated strict flag from @hydra.main and the Compose API (#1010)

Passing a config name as config_path to @hydra.main is now an error (#1010)

Promote hydra.experimental.{compose,initialize,initialize_config_dir,initialize_config_module} out from the hydra.experimental module to the hydra module. (#1030)

ConfigSourcePlugins needs to be modified to support recursive defaults, see link for details (#1080)

Default composition order change, see this for details (#1217)

Instantiate is now recursive by default, use recursive=False to disable recursive instantiation. (#566)

Bug Fixes

Hydra no longer errorenously changes the USER environment variable in pytest unit tests once installed (#1059)

Allow @ symbol in unquoted values in overrides (#1074)

Fix a bug where Structured Config has a field like 'list : Optional[List[int]] = None` (#1117)

No longer modifies exception stack trace when running under a debugger (#1237)

Fixed hydra.job.id and hydra.job.num not getting passed to jobs in multirun (#1270)

Add support for %f directive (microseconds) to the ${now:} resolver (#1287)

Fix interpolation in Hydra config node (#1335)

Fixed a bug where tab completion did not work if the Defaults List had a missing (???) item. (#1381)

Fix add overrides to properly insert new values into Structured Configs. (#1515)

Plugins

Add Optuna Sweeper plugin

Add support for multi-objective optimization to Optuna sweeper

Improved Documentation

New Defaults List [page])(https://hydra.cc/docs/next/advanced/defaults_list) (#1170)

New Extending Configs pattern (#1170)

Major updates to the Packages page (#1170)

New Configuring Experiments pattern (#1170)

Source code(tar.gz)
Source code(zip)
v1.0.6(Feb 1, 2021)
1.0.6 (2021-01-29)

Bug Fixes

Fix interpolation in Hydra config node (#1335)

Maintenance Changes

Remove Plugins from Singleton.get_state() (#1330)

Source code(tar.gz)
Source code(zip)
v1.0.5(Jan 7, 2021)
1.0.5 (2021-01-07)

Features

Support Python 3.9 . (#1062)

API Change (Renames, deprecations and removals)

Deprecate support for renaming packages via the command line. (#1140)

Bug Fixes

Fixed hydra.job.id and hydra.job.num not getting passed to jobs in multirun (#1270)

Plugins

Support additional_parameters as an optional param in the Submitit launcher plugin.

Add Optuna Sweeper plugin

Maintenance Changes

Limit OmegaConf depedency to 2.0 versions (#1253)

Source code(tar.gz)
Source code(zip)
v1.0.4(Nov 18, 2020)
1.0.4 (2020-11-17)

Bug Fixes

Hydra no longer errorenously changes the USER environment variable in pytest unit tests once installed (#1059)

Allow @ symbol in unquoted values in overrides (#1074)

Fix a bug where Structured Config has a field like 'list : Optional[List[int]] = None` (#1117)

Plugins

Switch Ax Sweeper plugin to use new Hydra command line syntax

Switch Nevergrad Sweeper plugin to use new Hydra command line syntax

Revert the use of "_" to separate the path element in keys for the Ax sweeper plugin.

Source code(tar.gz)
Source code(zip)
v1.0.3(Sep 23, 2020)
1.0.3 (2020-09-22)

Bug Fixes

Fixed interaction between interpolation and instantiate (#1001)

Fix float parsing for values where the exponent starts with 0 (e.g 1e-05) (#999)

Source code(tar.gz)
Source code(zip)
v1.0.2(Sep 20, 2020)
1.0.2 (2020-09-20)

This is a small update improving support for config dataclasses generated by configen.

Features

Overridden parameters annotated as Any in the config are not type checked by instantiate() (#982)

Source code(tar.gz)
Source code(zip)
v1.0.1(Sep 15, 2020)
1.0.1 (2020-09-15)

Features

Add ability to escape special characters in unquoted command line values (#954)

Bug Fixes

Add support for merging dicts and lists specified in the commend line (#939)

Fix appending dict values with new fields to config with + (#946)

Fix for Override Grammar integer parsing: Integers are now compatible with Python integers ('+' prefix, correct handling of '_') (#954)

Maintenance Changes

Upgrade to OmegaConf 2.0.2 (#755)

Source code(tar.gz)
Source code(zip)
v1.0.0(Sep 3, 2020)
1.0.0 (2020-09-03)

Hydra 1.0 is out! See blog post for high level details.

Major new features in Hydra 1.0

Config type safety via Structured Configs

More powerful command line

New plugins enabling remote launching and hyper parameter optimization

Improved error reporting

Reduce nesting levels with config packages

These release notes are covering the changes since 1.0.0rc4.

To see the complete release notes for 1.0 take a look at the release notes of:

1.0.0rc1

1.0.0rc2

1.0.0rc3

1.0.0rc4

1.0.0 (just below this line)

Features

Add support for casting interval to ints and floats (#915)

Bug Fixes

Fix to flush logging instead of shutting it down at job end (#833)

compose() no longer initialized logging subsystem (#833)

Fix overriding of hydra.job.env_set from the command line (#854)

Fix tab completion to not output errors in some cases when using --multirun (#856)

Fix OmegaConf custom resolvers not being propagated to Sweepers properly (#861)

Interpolation accepts any char between ${ and } (#891)

Update the signature of hydra.utils.instantiate() allow input configs of Any type (#896)

Fix instantiate to work correctly when parameters are interpolations into a parent node (#904)

Apps will now raise underlying exception if env HYDRA_FULL_ERROR=1 (#926)

Plugins

Fix integer/float choice from spec in the Nevergrad plugin

Improved Documentation

New page in Structured Configs tutorial: Static schema with many configs (#918)

New page in Structured Configs tutorial: Dynamic schema with many configs (#918)

Source code(tar.gz)
Source code(zip)
v1.0.0rc4(Aug 18, 2020)
1.0.0rc4 (2020-08-18)

Features

Upgrade to OmegaConf 2.0.1rc12 (#889)

Bug Fixes

Fix bug with bash autocompletion for the users who has . in their PATH environment variable (#868)

Fix instantiate to return None when None config is passed (#887)

Source code(tar.gz)
Source code(zip)
v1.0.0rc3(Aug 17, 2020)
1.0.0rc3 (2020-08-17)

Features

Extended Override grammar supports range(),glob(),type casts,shuffle, sort and more. (#752)

Add support for specifying an additional config dir in the command line (#874)

Allow overriding of the defaults and hydra nodes in primary Structured Configs that does not mentioned defaults and hydra. (#877)

API Change (Renames, deprecations and removals)

ObjectConf is deprecated for TargetConf. See upgrade page. (#882)

All plugins updated to use TargetConf. Be sure to update all your Hydra plugins (#882)

Instantiated objects no longer have a params node in the conf. keys like hydra.launcher.params.foo are changed to hydra.launcher.foo. Update your configs and overrides if you were overriding such parameters. (#882)

Bug Fixes

Fixed parsing of two nested lists with whitespace between them: [[a], [b]] (#836)

Improved Documentation

New Extended command line syntax page (#752)

Source code(tar.gz)
Source code(zip)
v1.0.0rc1(Aug 18, 2020)
1.0.0-rc1 (2020-05-31)

Hydra 1.0 is a major release introducing many new features and breaking some compatibility.

Features

Upgrade to OmegaConf 2.0 (Release notes) (#630)

Optional config type safety via Structured Configs (#629)

Improve command line and config composition error reporting (#349)

Hydra config can now be accessed through interpolation using ${hydra:key}, for example ${hydra:job.name} (#325)

Support for setting environment variable of running job (#7)

Changes command line processing (requiring + and ~ prefixes for appending and removing items) (#598)

Introducing @package header for config files (#586)

Add command line override flags for config_path and config_name (--config-path and --config-name) (#386)

hydra.main() now take an optional cfg object to passthrough to the function (#575)

Add hydra.experimental.{initialize_with_file, initialize_with_module} (#574)

Support for disabling the creation of the .hydra subdirectory by overriding "hydra.output_subdir" to "null" (#324)

Add hydra.utils.call() to call methods and functions as well as instantiate objects. Search module paths more generically. (#498)

Add support for overriding package from command line and defaults list (#235)

Config source is now abstracted, allowing additional config sources to be used (#257)

New ConfigSource plugin API allowing configs to be provided by external plugins (#367)

Add isort to ensure imports are always sorted (#340)

Codebase is now passing mypy --strict type checking (#342)

Improve performance of plugin discovery and instantiation (#489)

Modules whose name starts with "_" are skipped during plugin discovery (#494)

Plugins

Add Ax Sweeper plugin (Shagun Sodhani)

Add Nevergrad Sweeper plugin (Jérémy Rapin)

Add Joblib Launcher plugin (Jan-Matthis)

Add Submitit Launcher plugin to launch jobs to SLURM clusters

Add Fish shell Tab Completion plugin (Binsheng Liu) (#549)

API Change (Renames, deprecations and removals)

Drop support Python 2.7 and 3.5 (#313)

hydra.main() now takes an additional optional config_name and composite-style config_path is deprecated (#395)

Launcher API launch method now takes an additional initial_job_idx indicating the id of the first job in the batch (#284)

Singleton metaclass is now exposed at hydra.core.Singleton (#371)

Moved HydraConfig from hydra.plugins.common.utils to hydra.core (#371)

Move several formerly internal APIs to hydra/core to ensure plugins does not need to use internal APIs (#371)

Plugin import now requires explicit name (from hydra.plugins.launcher import Launcher) (#371)

Object Config "class" field is deprecated in favor of "cls" and will be removed in a future version. (#389)

Experimental compose API config_file changed to config_name (#395)

User plugins should be modified to not import twice during plugin discovery. see issue for details. (#482)

Change hydra.core.plugins.Plugins class to a Singleton. access should be changed to the pattern Plugins.instance().foo() (#489)

Plugins should now include test fixtures (sweep_runner, task_runner) via a standardized conftest.py (#521)

Switch Python 3 native namespace packages for plugins (See task for details) (#534)

Packaged configuration directories now requires an __init__.py at their top level (#536)

Appending config groups to the defaults list via the command line now requires a + prefix (#598)

Removing an item from the defaults list by assigning null (db=null) is deprecated, use ~db instead (#598)

Installed Hydra applications no longer need have an additional entry() function on the stack (#92)

Bug Fixes

Fix a bug causing sys.exit() error code to not be propagated (#351)

Shutdown logging subsystem aftter job finishes to ensure log files are flushed and closed (#378)

Fix a bug with utils.instantiate() failing if params contains interpolated values. (#388)

Allow hydra.utils.instantiate() to accept non primitive objects for passthrough by name (#400)

Fix to work when an Hydra app is executed in Jupyter notebook using the %run command (#481)

Plugins are no longer imported twice during plugin discovery (#482)

to_absolute_dir(path) now converts relative path to be relative to os.cwd() when used outside of Hydra (#496)

Improved Documentation

Working examples are provided for all Hydra plugins in plugins/examples (#253)

The basic tutorial was rewritten to reflect many changes (#602)

Add a new tutorial covering Structured Configs (#628)

Source code(tar.gz)
Source code(zip)
v1.0.0rc2(Jul 18, 2020)
1.0.0rc2 (2020-07-18)

Features

Introducing a DSL for the command line override syntax (#797)

Sanitize exception messages by default, showing only interesting stack frames (#646)

Add a new --info flag to show Hydra debug information without running the user function. (#662)

Add --package flag, can be used with --cfg to print a specific config package (#674)

Improve the Compose API initialization methods (#702)

Switch internal package resource loader from pkg_resources to importlb_resources (#719)

Allow interpolation in the Hydra config node to access the application config (#771)

Hydra provides pytest fixtures (hydra_sweep_runner, hydra_task_runner, hydra_restore_singletons) (#687)

Added contexts to help initialize Hydra in a localized context (Mostly for Unit Tests and Jupyter Notebooks) (#687)

Prints error when command line override looks like a sweep but --multirun is not specified (#712)

Frozen Structured Configs are respected during composition command line overrides (#724)

API Change (Renames, deprecations and removals)

Hydra argparse flags are now consistently using dash-style. (#661)

Restructure the configuration of the Submitit launcher (#682)

Deprecated cls in favor of target in hydra.utils.{instantiate()/call{}} (#721)

Bug Fixes

Add hydra/py.typed to pip package to enable type checking in downstream projects (#657)

Fix initilzation of Hydra in a module imported from a Jupyter Notebook (#695)

Fix handling of main module override via HYDRA_MAIN_MODULE (#717)

Fix interpoaltions in overrides to evaluate lazily (#725)

Fixed handling of multirun when target config node is introduced by composition (#726)

Sweepers are now validating config composition before launching (#775)

Save .hydra/*.yaml before task execution (#678)

Plugins

Add Redis Queue Launcher plugin (Jan-Matthis)

Upgrade ax-platform used by Ax sweeper plugin to 0.1.13 or newer

Improved Documentation

Add a unit testing example in the Hydra example application (#687)

New Terminology page (#795)

New Advanced/Hydra in Jupyter Notebooks page (#795)

New Advanced/Hydra in Unit Tests page (#795)

New Advanced/Command-line syntax page (#795)

Maintenance Changes

Integrate yamllint (#693)

Upgrade to OmegaConf 2.0.1rc11 which solves multiple bugs (#755)

Source code(tar.gz)
Source code(zip)
v0.11.3(Aug 18, 2020)
0.11.3 (2019-12-29)

Bug Fixes

Pin Hydra 0.11 to OmegaConf 1.4 to avoid breaking compatibility when OmegaConf 2.0 is released (#334)

Improved Documentation

Document a simple Ray example (#317)

Source code(tar.gz)
Source code(zip)