1052 Upgrade Ignite dependency to 0.4.2

[Nic-Ma]

Fixes #1052 .

Description

Ignite released v0.4.2 today, this PR upgraded it.

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh --codeformat --coverage.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[Nic-Ma]

Hi @ericspod and @wyli ,

This PR updated all the ignite related logic and hack to v0.4.2.
I already tested locally with the workflow and ignite examples, will submit another PR in tutorial repo to slightly update them.
Could you please help review it?

Thanks.

[wyli]

I think this should be held til after v0.3 release, unless we are fixing some critical bugs in the workflow

[Nic-Ma]

I think this should be held til after v0.3 release, unless we are fixing some critical bugs in the workflow

Sure, what are the critical bugs you mean?
Thanks.

[wyli]

I think this should be held til after v0.3 release, unless we are fixing some critical bugs in the workflow

Sure, what are the critical bugs you mean?
Thanks.

meaning things that would break the software features that we'd like to highlight in v0.3

[Nic-Ma]

Found some distributed training issue in ignite v0.4.2, mark as WIP.
Thanks.

[Nic-Ma]

Tracking the distributed training issue in: pytorch/ignite#1307
Thanks.

[Nic-Ma]

Tracking the distributed training issue in: pytorch/ignite#1307
Thanks.

issue fixed.

[Nic-Ma]

Hi @wyli ,

I updated the workflow integration test and verified locally.
It's OK for review now.

Thanks.

[wyli]

thanks @Nic-Ma may want to incorporate the updated API with a deterministic flag:
https://github.com/pytorch/ignite/blob/84b5d413b68cf0407cb7bd027bec194809a5db2c/ignite/engine/__init__.py#L41-L50

[Nic-Ma]

thanks @Nic-Ma may want to incorporate the updated API with a deterministic flag:
https://github.com/pytorch/ignite/blob/84b5d413b68cf0407cb7bd027bec194809a5db2c/ignite/engine/__init__.py#L41-L50

I think we prefer to use monai.utils.set_determinism() for deterministic training, so I didn't add it to our multi_gpu_trainer.
Thanks.

[rijobro]

Running one of the tutorials, I now get the following error:

Click for error

---------------------------------------------------------------------------
OptionalImportError                       Traceback (most recent call last)
<ipython-input-6-8f07429d3c99> in <module>
     11 from monai.config import print_config
     12 from monai.data import CacheDataset, DataLoader
---> 13 from monai.engines import GanKeys, GanTrainer, default_make_latent
     14 from monai.handlers import CheckpointSaver, MetricLogger, StatsHandler
     15 from monai.networks import normal_init

/opt/monai/monai/engines/__init__.py in <module>
     10 # limitations under the License.
     11 
---> 12 from .evaluator import *
     13 from .multi_gpu_supervised_trainer import *
     14 from .trainer import *

/opt/monai/monai/engines/evaluator.py in <module>
     17 from monai.engines.utils import CommonKeys as Keys
     18 from monai.engines.utils import default_prepare_batch
---> 19 from monai.engines.workflow import Workflow
     20 from monai.inferers import Inferer, SimpleInferer
     21 from monai.transforms import Transform

/opt/monai/monai/engines/workflow.py in <module>
     31 
     32 
---> 33 class Workflow(IgniteEngine):  # type: ignore[valid-type, misc] # due to optional_import
     34     """
     35     Workflow defines the core work process inheriting from Ignite engine.

/opt/monai/monai/utils/module.py in __getattr__(self, name)
    195                 OptionalImportError: When you call this method.
    196             """
--> 197             raise self._exception
    198 
    199         def __call__(self, *_args, **_kwargs):

OptionalImportError: from ignite.engine import Engine (requires 'ignite.engine 0.3.0' by 'exact_version').

I'm running MONAI master and ignite 0.4.2, as verified with print_config:

MONAI version: 0.3.0rc3+9.gc009938
Python version: 3.8.3 (default, May 19 2020, 18:47:26)  [GCC 7.3.0]
OS version: Linux (4.15.0-91-generic)
Numpy version: 1.19.2
Pytorch version: 1.6.0

Optional dependencies:
Pytorch Ignite version: 0.4.2
Nibabel version: 3.1.1
scikit-image version: 0.17.2
Pillow version: 7.2.0
Tensorboard version: 2.3.0
gdown version: 3.12.2
TorchVision version: 0.7.0
ITK version: 5.1.1
tqdm version: 4.50.0

[wyli]

Running one of the tutorials, I now get the following error:

Click for error

---------------------------------------------------------------------------
OptionalImportError                       Traceback (most recent call last)
<ipython-input-6-8f07429d3c99> in <module>
     11 from monai.config import print_config
     12 from monai.data import CacheDataset, DataLoader
---> 13 from monai.engines import GanKeys, GanTrainer, default_make_latent
     14 from monai.handlers import CheckpointSaver, MetricLogger, StatsHandler
     15 from monai.networks import normal_init

/opt/monai/monai/engines/__init__.py in <module>
     10 # limitations under the License.
     11 
---> 12 from .evaluator import *
     13 from .multi_gpu_supervised_trainer import *
     14 from .trainer import *

/opt/monai/monai/engines/evaluator.py in <module>
     17 from monai.engines.utils import CommonKeys as Keys
     18 from monai.engines.utils import default_prepare_batch
---> 19 from monai.engines.workflow import Workflow
     20 from monai.inferers import Inferer, SimpleInferer
     21 from monai.transforms import Transform

/opt/monai/monai/engines/workflow.py in <module>
     31 
     32 
---> 33 class Workflow(IgniteEngine):  # type: ignore[valid-type, misc] # due to optional_import
     34     """
     35     Workflow defines the core work process inheriting from Ignite engine.

/opt/monai/monai/utils/module.py in __getattr__(self, name)
    195                 OptionalImportError: When you call this method.
    196             """
--> 197             raise self._exception
    198 
    199         def __call__(self, *_args, **_kwargs):

OptionalImportError: from ignite.engine import Engine (requires 'ignite.engine 0.3.0' by 'exact_version').

I'm running MONAI master and ignite 0.4.2, as verified with print_config:

MONAI version: 0.3.0rc3+9.gc009938
Python version: 3.8.3 (default, May 19 2020, 18:47:26)  [GCC 7.3.0]
OS version: Linux (4.15.0-91-generic)
Numpy version: 1.19.2
Pytorch version: 1.6.0

Optional dependencies:
Pytorch Ignite version: 0.4.2
Nibabel version: 3.1.1
scikit-image version: 0.17.2
Pillow version: 7.2.0
Tensorboard version: 2.3.0
gdown version: 3.12.2
TorchVision version: 0.7.0
ITK version: 5.1.1
tqdm version: 4.50.0

perhaps there are multiple copies of monai installed on your system? the last line of the error "(requires 'ignite.engine 0.3.0' by 'exact_version')" is not from 0.3.0rc3+9.gc009938

[rijobro]

Yeah I think you're right, sorry about that.

[wyli]

Yeah I think you're right, sorry about that.

no problem, we'll tag 0.3.0rc4 soon, should be non-breaking since 0.3.0rc2