update features

examples/ae.py

@@ -108,7 +108,7 @@ def __init__(self, hparams: HParams) -> None:
             "loss": MeanMetric(),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.encoder = encoder
         self.decoder = decoder
         self.lr_s = lr_s

examples/cl.py

@@ -67,7 +67,7 @@ def __init__(self, hparams: HParams) -> None:
         }
         self.temperature = hparams.temperature
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.resnet = resnet
         self.lr_s = lr_s
 

examples/cl_ddp.py

@@ -71,7 +71,7 @@ def __init__(self, hparams: HParams) -> None:
         }
         self.temperature = hparams.temperature
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.resnet = resnet
         self.lr_s = lr_s
 

examples/cl_ddp_spawn.py

@@ -72,7 +72,7 @@ def __init__(self, hparams: HParams) -> None:
         }
         self.temperature = hparams.temperature
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.resnet = resnet
         self.lr_s = lr_s
 

examples/cv.py

@@ -54,7 +54,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("multiclass", num_classes=num_classes),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss()

examples/cv_ddp.py

@@ -59,7 +59,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("multiclass", num_classes=num_classes),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss()

examples/cv_ddp_spawn.py

@@ -59,7 +59,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("multiclass", num_classes=num_classes),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss()

examples/dqn.py

@@ -149,7 +149,7 @@ def get_rand_p(global_step: int, T_max: int, eta_min: float, eta_max: float) ->
 
 class MyLModule(ml.LModule):
     def __init__(self, memo_pool: MemoryPool, hparams: HParams) -> None:
-
+        self.hparams: HParams
         env = gym.make(hparams.env_name, render_mode=RENDER_MODE)
         #
         in_channels: int = env.observation_space.shape[0]
@@ -158,7 +158,7 @@ def __init__(self, memo_pool: MemoryPool, hparams: HParams) -> None:
         agent = Agent(env, memo_pool, model, ml.select_device(device_ids))
 
         optimizer = getattr(optim, hparams.optim_name)(model.parameters(), **hparams.optim_hparams)
-        super().__init__([optimizer], {}, hparams.__dict__)
+        super().__init__([optimizer], {}, hparams)
         self.model = model
         self.old_model = deepcopy(self.model).requires_grad_(False)
         # New_model and old_model are used for model training.
@@ -169,10 +169,10 @@ def __init__(self, memo_pool: MemoryPool, hparams: HParams) -> None:
         self.agent = agent
         self.get_rand_p = partial(get_rand_p, **hparams.rand_p)
         #
-        self.warmup_memory_steps = self.hparams["warmup_memory_steps"]
+        self.warmup_memory_steps = self.hparams.warmup_memory_steps
         # synchronize the model every sync_steps
-        self.sync_steps = self.hparams["sync_steps"]
-        self.gamma = self.hparams["gamma"]  # reward decay
+        self.sync_steps = self.hparams.sync_steps
+        self.gamma = self.hparams.gamma  # reward decay
 
         #
         self._warmup_memo(self.warmup_memory_steps)

examples/gan.py

@@ -101,7 +101,7 @@ def __init__(self, hparams: HParams) -> None:
         G, D = Generator(**hparams.G_hparams), Discriminator(**hparams.D_hparams)
         opt_G = getattr(optim, hparams.opt_G_name)(G.parameters(), **hparams.opt_G_hparams)
         opt_D = getattr(optim, hparams.opt_D_name)(D.parameters(), **hparams.opt_D_hparams)
-        super().__init__([opt_G, opt_D], {}, hparams.__dict__)
+        super().__init__([opt_G, opt_D], {}, hparams)
         self.G = G
         self.D = D
         self.loss_fn = nn.BCEWithLogitsLoss()

examples/gnn.py

@@ -107,7 +107,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("multiclass", num_classes=num_classes),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss()

examples/gnn2.py

@@ -87,7 +87,7 @@ def __init__(self, hparams: HParams) -> None:
             "auc": AUROC("binary")
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.BCEWithLogitsLoss()

examples/gnn3.py

@@ -93,7 +93,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("binary"),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.BCEWithLogitsLoss()

examples/meta_learning.py

@@ -132,7 +132,7 @@ def __init__(self, hparams: HParams) -> None:
             "acc":  Accuracy("multiclass", num_classes=proto_dim),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss()

examples/nlp.py

@@ -60,7 +60,7 @@ def __init__(self, hparams: HParams) -> None:
         }
         lr_s: LRScheduler = lrs.CosineAnnealingLR(optimizer, **hparams.lrs_hparams)
         lr_s = ml.warmup_decorator(lr_s, hparams.warmup)
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.model = model
         self.lr_s = lr_s
         self.loss_fn = nn.CrossEntropyLoss(label_smoothing=0.1)

examples/vae.py

@@ -155,7 +155,7 @@ def __init__(self, hparams: HParams) -> None:
             "loss": MeanMetric(),
         }
         #
-        super().__init__([optimizer], metrics, hparams.__dict__)
+        super().__init__([optimizer], metrics, hparams)
         self.mse = nn.MSELoss()
         self.encoder = encoder
         self.decoder = decoder

mini_lightning/_mini_lightning.py

@@ -20,17 +20,18 @@ def __init__(
         self,
         optimizers: List[Optimizer],
         metrics: Dict[str, Metric],
-        hparams: Optional[Dict[str, Any]] = None
+        hparams: Any = None
     ) -> None:
         """
         optimizers: Use List, for supporting GAN
         hparams: Hyperparameters to be saved
+            object or Dict[str, Any] or None
         """
         # _models: for trainer_init(device, ddp), _epoch_start(train, eval); print_model_info; save_ckpt
         self._models: List[str] = []
         self.optimizers = optimizers
         self.metrics = metrics
-        self.hparams: Dict[str, Any] = hparams if hparams is not None else {}
+        self.hparams = hparams
         self.trainer: Optional["Trainer"] = None
 
     @property
@@ -516,9 +517,13 @@ def _check_hparams(self, hparams: Any) -> Any:
             res = repr(hparams)  # e.g. function
         return res
 
-    def save_hparams(self, hparams: Dict[str, Any]) -> None:
+    def save_hparams(self, hparams: Any) -> None:
         if self.rank not in {-1, 0}:
             return
+        if hparams is None:
+            hparams = {}
+        elif not isinstance(hparams, dict):
+            hparams = hparams.__dict__
         saved_hparams = self._check_hparams(hparams)
         logger.info(f"Saving hparams: {saved_hparams}")
         write_to_yaml(saved_hparams, self.hparams_path)

setup.py

@@ -21,7 +21,7 @@ def read_file(path: str) -> str:
 ]
 setup(
     name="mini-lightning",
-    version="0.1.10",
+    version="0.2.0",
     description=description,
     long_description=long_description,
     long_description_content_type="text/markdown",