Fix python 3.7 compatibility bugs (#233)

- Limit pytorch version to 1.8.X, from >= 1.8: PyTorch suddenly got
  really picky about type annotations on subclasses of IterableDataset
  for some unexplained reason, and that breaks most of the
  EnvDataset-related wrappers in Sequoia.

- Fix compatibility issues with `lru_cache` taking one required argument
  in python 3.7

- Fix PyTorch-Lightning==1.4.0 compatibility issues due to ModelSummary
  not having a MODE_DEFAULT anymore

- Fix `typing.get_origin` bug with python 3.7 in typed_dict.py

Signed-off-by: Fabrice Normandin <fabrice.normandin@gmail.com>

requirements.txt

@@ -6,7 +6,10 @@ nngeometry @ git+https://github.com/oleksost/nngeometry.git#egg=nngeometry
 pyyaml!=5.4.*,>=5.1
 simple_parsing>=0.0.15.post1
 matplotlib==3.2.2
-torch>=1.8
+# NOTE: @lebrice: PyTorch suddenly got really picky about type annotations in 1.9.0 for
+# some reason, and they really don't do a great job at evaluating them, so removing it
+# for now.
+torch>=1.8,<1.9.0
 torchvision>=0.9
 scikit-learn
 tqdm
@@ -17,7 +20,7 @@ plotly
 pandas
 # Only for python < 3.8
 singledispatchmethod;python_version<'3.8'
-# Temporarily fix the pytorch lightning version (issue #134)
+# NOTE: PyTorch-Lightning version 1.4.0 is "working" but raises lots of warnings.
 pytorch-lightning>=1.3.8
 pytorch-lightning-bolts>=0.3.2
 # Requirements for running tests:

sequoia/common/gym_wrappers/env_dataset.py

@@ -6,6 +6,7 @@
     Callable,
     Dict,
     Generator,
+    Iterator,
     Generic,
     Iterable,
     List,
@@ -224,7 +225,7 @@ def send(self, action: ActionType) -> RewardType:
         self.n_sends_ += 1
         return self.reward_
 
-    def __iter__(self) -> Iterable[ObservationType]:
+    def __iter__(self) -> Iterator[ObservationType]:
         """Iterator for an episode in the environment, which uses the 'active
         dataset' style with __iter__ and send.
 

sequoia/common/gym_wrappers/policy_env.py

@@ -4,7 +4,17 @@
 This policy should then accept the 'state' or something like that.
 """
 from dataclasses import dataclass
-from typing import Any, Callable, Iterable, Optional, Tuple, TypeVar, Dict, Generic
+from typing import (
+    Any,
+    Callable,
+    Iterator,
+    Optional,
+    Tuple,
+    TypeVar,
+    Dict,
+    Generic,
+    Iterable,
+)
 
 import gym
 from torch.utils.data import IterableDataset
@@ -169,7 +179,7 @@ def reset(self, *args, **kwargs) -> None:
         self._n_steps_in_episode = 0
         return self._observation
 
-    def __iter__(self) -> Iterable[DatasetItem]:
+    def __iter__(self) -> Iterator[DatasetItem]:
         """Iterator for an episode/trajectory in the env.
         
         This uses the policy to iteratively perform an episode in the env, and

sequoia/common/spaces/typed_dict.py

@@ -43,6 +43,13 @@
 S = TypeVar("S")
 Dataclass = TypeVar("Dataclass")
 
+try:
+    from typing import get_origin
+except ImportError:
+    # Python 3.7's typing module doesn't have this `get_origin` function, so get it from
+    # `typing_inspect`.
+    from typing_inspect import get_origin
+
 
 class TypedDictSpace(spaces.Dict, Mapping[str, Space], Generic[M]):
     """ Subclass of `spaces.Dict` that allows custom dtypes and uses type annotations.
@@ -183,7 +190,7 @@ def __init__(
                 if isclass(type_annotation) and issubclass(type_annotation, gym.Space):
                     is_space = True
                 else:
-                    origin = typing.get_origin(type_annotation)
+                    origin = get_origin(type_annotation)
                     is_space = (
                         origin is not None
                         and isclass(origin)

sequoia/methods/__init__.py

@@ -93,7 +93,7 @@ def wrap(method_class: Type[Method]) -> Type[Method]:
     return wrap(method_class)
 
 
-@lru_cache
+@lru_cache(1)
 def get_external_methods() -> Dict[str, Type[Method]]:
     """ Returns a dictionary of the Methods defined outside of Sequoia.
 

sequoia/methods/models/base_model/model.py

@@ -748,10 +748,10 @@ def shared_modules(self) -> Dict[str, nn.Module]:
             shared_modules["output_head"] = self.output_head
         return shared_modules
 
-    def summarize(self, mode: str = ModelSummary.MODE_DEFAULT) -> ModelSummary:
-        model_summary = ModelSummary(self, mode=mode)
-        log.debug("\n" + str(model_summary))
-        return model_summary
+    # def summarize(self, mode: str = ModelSummary.MODE_DEFAULT) -> ModelSummary:
+    #     model_summary = ModelSummary(self, mode=mode)
+    #     log.debug("\n" + str(model_summary))
+    #     return model_summary
 
     def _are_batched(self, observations: IncrementalAssumption.Observations) -> bool:
         """ Returns wether these observations are batched. """

sequoia/settings/rl/wrappers/measure_performance.py

@@ -20,7 +20,9 @@
 
 
 class MeasureRLPerformanceWrapper(
-    MeasurePerformanceWrapper[ActiveEnvironment, EpisodeMetrics]
+    MeasurePerformanceWrapper
+    # MeasurePerformanceWrapper[ActiveEnvironment]  # python 3.7
+    # MeasurePerformanceWrapper[ActiveEnvironment, EpisodeMetrics] # python 3.8+
 ):
     def __init__(
         self,

sequoia/settings/sl/continual/setting.py

@@ -184,7 +184,7 @@ class ContinualSLSetting(SLSetting, ContinualAssumption):
 
     def __post_init__(self):
         super().__post_init__()
-        assert not self.has_setup_fit
+        # assert not self.has_setup_fit
         # Test values default to the same as train.
         self.test_increment = self.test_increment or self.increment
         self.test_initial_increment = (

sequoia/settings/sl/setting.py

@@ -10,7 +10,7 @@
 from sequoia.settings import Setting
 from sequoia.common.transforms import Transforms
 from .environment import PassiveEnvironment
-from torch.tensor import Tensor
+from torch import Tensor
 
 
 @dataclass

sequoia/settings/sl/wrappers/measure_performance.py

@@ -12,7 +12,7 @@
 TODO: Move this somewhere more appropriate. There's also the RL version of the wrapper
 here.
 """
-from typing import Any, Dict, Generic, Iterable, List, Optional, Sequence, Tuple, Union
+from typing import Any, Dict, Generic, Iterator, List, Optional, Sequence, Tuple, Union
 
 import numpy as np
 import wandb
@@ -30,7 +30,9 @@
 
 
 class MeasureSLPerformanceWrapper(
-    MeasurePerformanceWrapper[PassiveEnvironment, ClassificationMetrics]
+    MeasurePerformanceWrapper,
+    # MeasurePerformanceWrapper[PassiveEnvironment]  # Python 3.7
+    # MeasurePerformanceWrapper[PassiveEnvironment, ClassificationMetrics] # Python 3.8+
 ):
     def __init__(
         self,
@@ -139,7 +141,7 @@ def get_metrics(self, action: Actions, reward: Rewards) -> Metrics:
             wandb.log(log_dict)
         return metric
 
-    def __iter__(self) -> Iterable[Tuple[Observations, Optional[Rewards]]]:
+    def __iter__(self) -> Iterator[Tuple[Observations, Optional[Rewards]]]:
         if self.__epochs == 1 and self.first_epoch_only:
             print(
                 colorize(