Cast hparams to dict when not using omegaconf (#4770)

* init fix

* init test

* more specific dict assert

* update changelog

* Update tests/checkpointing/test_model_checkpoint.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

Co-authored-by: chaton <thomas@grid.ai>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

CHANGELOG.md

@@ -60,6 +60,29 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
 
+## [unreleased.BugFix] - YYYY-MM-DD
+
+### Added
+
+
+
+### Changed
+
+
+
+### Deprecated
+
+
+
+### Removed
+
+
+
+### Fixed
+
+- Fixed checkpoint hparams dict casting when omegaconf is available ([#4770](https://github.com/PyTorchLightning/pytorch-lightning/pull/4770))
+
+
 ## [1.0.7] - 2020-11-17
 
 ### Added

pytorch_lightning/trainer/connectors/checkpoint_connector.py

@@ -328,10 +328,9 @@ def dump_checkpoint(self, weights_only: bool = False) -> dict:
             if hasattr(model, '_hparams_name'):
                 checkpoint[LightningModule.CHECKPOINT_HYPER_PARAMS_NAME] = model._hparams_name
             # dump arguments
-            if OMEGACONF_AVAILABLE:
+            if OMEGACONF_AVAILABLE and isinstance(model.hparams, Container):
                 checkpoint[LightningModule.CHECKPOINT_HYPER_PARAMS_KEY] = model.hparams
-                if isinstance(model.hparams, Container):
-                    checkpoint[LightningModule.CHECKPOINT_HYPER_PARAMS_TYPE] = type(model.hparams)
+                checkpoint[LightningModule.CHECKPOINT_HYPER_PARAMS_TYPE] = type(model.hparams)
             else:
                 checkpoint[LightningModule.CHECKPOINT_HYPER_PARAMS_KEY] = dict(model.hparams)
 

tests/checkpointing/test_model_checkpoint.py

@@ -27,6 +27,8 @@
 import cloudpickle
 import pytest
 import torch
+from omegaconf import Container, OmegaConf
+from argparse import Namespace
 
 import tests.base.develop_utils as tutils
 from pytorch_lightning import Trainer, seed_everything
@@ -911,3 +913,35 @@ def training_step(self, *args):
     expected = float("inf" if mode == "min" else "-inf")
     assert model_checkpoint.best_model_score == expected
     assert model_checkpoint.current_score == expected
+
+
+@pytest.mark.parametrize("hparams_type", [dict, Container])
+def test_hparams_type(tmpdir, hparams_type):
+    class TestModel(BoringModel):
+        def __init__(self, hparams):
+            super().__init__()
+            self.save_hyperparameters(hparams)
+
+    model_checkpoint = ModelCheckpoint(
+        dirpath=tmpdir,
+        save_top_k=1,
+        monitor="foo",
+    )
+    trainer = Trainer(
+        default_root_dir=tmpdir,
+        fast_dev_run=True,
+        callbacks=[model_checkpoint],
+        logger=False,
+        weights_summary=None,
+        progress_bar_refresh_rate=0,
+    )
+    hp = {"test_hp_0": 1, "test_hp_1": 2}
+    hp = OmegaConf.create(hp) if hparams_type == Container else Namespace(**hp)
+    model = TestModel(hp)
+    trainer.fit(model)
+    ckpt = trainer.checkpoint_connector.dump_checkpoint()
+    if hparams_type == Container:
+        assert isinstance(ckpt[model.CHECKPOINT_HYPER_PARAMS_KEY], hparams_type)
+    else:
+        # make sure it's not AttributeDict
+        assert type(ckpt[model.CHECKPOINT_HYPER_PARAMS_KEY]) == hparams_type