Merge branch 'master' into feat-sync_step

Author: borisdayma

.drone.yml

@@ -30,6 +30,7 @@ steps:
     MKL_THREADING_LAYER: GNU
 
   commands:
+    - set -e
     - python --version
     - pip --version
     - nvidia-smi

.github/workflows/ci_test-base.yml

@@ -1,4 +1,4 @@
-name: CI base testing
+name: CI basic testing
 
 # see: https://help.github.com/en/actions/reference/events-that-trigger-workflows
 on:  # Trigger the workflow on push or pull request, but only for the master branch

CHANGELOG.md

@@ -5,7 +5,7 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
-## [1.1.3rc] - 2020-12-29
+## [1.1.3] - 2021-01-05
 
 ### Added
 
@@ -25,12 +25,13 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Fixed
 
+- Skip restore from `resume_from_checkpoint` in while `testing` ([#5161](https://github.com/PyTorchLightning/pytorch-lightning/pull/5161))
+
 - Allowed `log_momentum` for adaptive optimizers in `LearningRateMonitor` ([#5333](https://github.com/PyTorchLightning/pytorch-lightning/pull/5333))
 
 - Disabled checkpointing, earlystopping and logger with `fast_dev_run` ([#5277](https://github.com/PyTorchLightning/pytorch-lightning/pull/5277))
 
 
-
 ## [1.1.2] - 2020-12-23
 
 ### Added

docs/source/transfer_learning.rst

@@ -52,16 +52,22 @@ Example: Imagenet (computer Vision)
 
     class ImagenetTransferLearning(LightningModule):
         def __init__(self):
+            super().__init__()
+
             # init a pretrained resnet
-            num_target_classes = 10
-            self.feature_extractor = models.resnet50(pretrained=True)
-            self.feature_extractor.eval()
+            backbone = models.resnet50(pretrained=True)
+            num_filters = backbone.fc.in_features
+            layers = list(backbone.children())[:-1]
+            self.feature_extractor = torch.nn.Sequential(*layers)
 
             # use the pretrained model to classify cifar-10 (10 image classes)
-            self.classifier = nn.Linear(2048, num_target_classes)
+            num_target_classes = 10
+            self.classifier = nn.Linear(num_filters, num_target_classes)
 
         def forward(self, x):
-            representations = self.feature_extractor(x)
+            self.feature_extractor.eval()
+            with torch.no_grad():
+                representations = self.feature_extractor(x).flatten(1)
             x = self.classifier(representations)
             ...
 

pytorch_lightning/callbacks/model_checkpoint.py

@@ -208,6 +208,7 @@ def on_save_checkpoint(self, trainer, pl_module) -> Dict[str, Any]:
             "best_model_score": self.best_model_score,
             "best_model_path": self.best_model_path,
             "current_score": self.current_score,
+            "dirpath": self.dirpath
         }
 
     def on_load_checkpoint(self, checkpointed_state: Dict[str, Any]):

pytorch_lightning/loggers/__init__.py

@@ -24,40 +24,25 @@
     'CSVLogger',
 ]
 
-try:
-    # needed to prevent ImportError and duplicated logs.
-    environ["COMET_DISABLE_AUTO_LOGGING"] = "1"
+from pytorch_lightning.loggers.comet import _COMET_AVAILABLE, CometLogger
+from pytorch_lightning.loggers.mlflow import _MLFLOW_AVAILABLE, MLFlowLogger
+from pytorch_lightning.loggers.neptune import _NEPTUNE_AVAILABLE, NeptuneLogger
+from pytorch_lightning.loggers.test_tube import _TESTTUBE_AVAILABLE, TestTubeLogger
+from pytorch_lightning.loggers.wandb import _WANDB_AVAILABLE, WandbLogger
 
-    from pytorch_lightning.loggers.comet import CometLogger
-except ImportError:  # pragma: no-cover
-    del environ["COMET_DISABLE_AUTO_LOGGING"]  # pragma: no-cover
-else:
+if _COMET_AVAILABLE:
     __all__.append('CometLogger')
+    # needed to prevent ImportError and duplicated logs.
+    environ["COMET_DISABLE_AUTO_LOGGING"] = "1"
 
-try:
-    from pytorch_lightning.loggers.mlflow import MLFlowLogger
-except ImportError:  # pragma: no-cover
-    pass  # pragma: no-cover
-else:
+if _MLFLOW_AVAILABLE:
     __all__.append('MLFlowLogger')
 
-try:
-    from pytorch_lightning.loggers.neptune import NeptuneLogger
-except ImportError:  # pragma: no-cover
-    pass  # pragma: no-cover
-else:
+if _NEPTUNE_AVAILABLE:
     __all__.append('NeptuneLogger')
 
-try:
-    from pytorch_lightning.loggers.test_tube import TestTubeLogger
-except ImportError:  # pragma: no-cover
-    pass  # pragma: no-cover
-else:
+if _TESTTUBE_AVAILABLE:
     __all__.append('TestTubeLogger')
 
-try:
-    from pytorch_lightning.loggers.wandb import WandbLogger
-except ImportError:  # pragma: no-cover
-    pass  # pragma: no-cover
-else:
+if _WANDB_AVAILABLE:
     __all__.append('WandbLogger')

pytorch_lightning/loggers/comet.py

@@ -21,17 +21,18 @@
 from argparse import Namespace
 from typing import Any, Dict, Optional, Union
 
-try:
-    import comet_ml
+import torch
+from torch import is_tensor
 
-except ModuleNotFoundError:  # pragma: no-cover
-    comet_ml = None
-    CometExperiment = None
-    CometExistingExperiment = None
-    CometOfflineExperiment = None
-    API = None
-    generate_guid = None
-else:
+from pytorch_lightning import _logger as log
+from pytorch_lightning.loggers.base import LightningLoggerBase, rank_zero_experiment
+from pytorch_lightning.utilities import rank_zero_only, _module_available
+from pytorch_lightning.utilities.exceptions import MisconfigurationException
+
+_COMET_AVAILABLE = _module_available("comet_ml")
+
+if _COMET_AVAILABLE:
+    import comet_ml
     from comet_ml import ExistingExperiment as CometExistingExperiment
     from comet_ml import Experiment as CometExperiment
     from comet_ml import OfflineExperiment as CometOfflineExperiment
@@ -41,14 +42,11 @@
     except ImportError:  # pragma: no-cover
         # For more information, see: https://www.comet.ml/docs/python-sdk/releases/#release-300
         from comet_ml.papi import API  # pragma: no-cover
-
-import torch
-from torch import is_tensor
-
-from pytorch_lightning import _logger as log
-from pytorch_lightning.loggers.base import LightningLoggerBase, rank_zero_experiment
-from pytorch_lightning.utilities import rank_zero_only
-from pytorch_lightning.utilities.exceptions import MisconfigurationException
+else:
+    # needed for test mocks, these tests shall be updated
+    comet_ml = None
+    CometExperiment, CometExistingExperiment, CometOfflineExperiment = None, None, None
+    API = None
 
 
 class CometLogger(LightningLoggerBase):

pytorch_lightning/loggers/mlflow.py

@@ -21,21 +21,25 @@
 from time import time
 from typing import Any, Dict, Optional, Union
 
-try:
-    import mlflow
-    from mlflow.tracking import MlflowClient
-except ModuleNotFoundError:  # pragma: no-cover
-    mlflow = None
-    MlflowClient = None
-
 
 from pytorch_lightning import _logger as log
 from pytorch_lightning.loggers.base import LightningLoggerBase, rank_zero_experiment
-from pytorch_lightning.utilities import rank_zero_only, rank_zero_warn
+from pytorch_lightning.utilities import rank_zero_only, rank_zero_warn, _module_available
+
 
 LOCAL_FILE_URI_PREFIX = "file:"
 
 
+_MLFLOW_AVAILABLE = _module_available("mlflow")
+try:
+    import mlflow
+    from mlflow.tracking import MlflowClient
+# todo: there seems to be still some remaining import error with Conda env
+except ImportError:
+    _MLFLOW_AVAILABLE = False
+    mlflow, MlflowClient = None, None
+
+
 class MLFlowLogger(LightningLoggerBase):
     """
     Log using `MLflow <https://mlflow.org>`_.

pytorch_lightning/loggers/neptune.py

@@ -17,21 +17,23 @@
 --------------
 """
 from argparse import Namespace
-from typing import Any, Dict, Iterable, List, Optional, Union
-
-try:
-    import neptune
-    from neptune.experiments import Experiment
-except ImportError:  # pragma: no-cover
-    neptune = None
-    Experiment = None
+from typing import Any, Dict, Iterable, Optional, Union
 
 import torch
 from torch import is_tensor
 
 from pytorch_lightning import _logger as log
 from pytorch_lightning.loggers.base import LightningLoggerBase, rank_zero_experiment
-from pytorch_lightning.utilities import rank_zero_only
+from pytorch_lightning.utilities import rank_zero_only, _module_available
+
+_NEPTUNE_AVAILABLE = _module_available("neptune")
+
+if _NEPTUNE_AVAILABLE:
+    import neptune
+    from neptune.experiments import Experiment
+else:
+    # needed for test mocks, these tests shall be updated
+    neptune, Experiment = None, None
 
 
 class NeptuneLogger(LightningLoggerBase):

pytorch_lightning/loggers/tensorboard.py

@@ -217,7 +217,7 @@ def save(self) -> None:
         hparams_file = os.path.join(dir_path, self.NAME_HPARAMS_FILE)
 
         # save the metatags file if it doesn't exist
-        if not os.path.isfile(hparams_file):
+        if not self._fs.isfile(hparams_file):
             save_hparams_to_yaml(hparams_file, self.hparams)
 
     @rank_zero_only