Dev (#136)

* comment improvements

* add __init__.py

* add disabling pin memory on fast dev run

* rename template_utils.py to utils.py

* improve README.md

* add num_classes property to mnist datamodule

* add sh to requirements.txt

* bump package versions

README.md

@@ -96,7 +96,7 @@ The directory structure of new project looks like this:
 │   ├── models                  <- Lightning models
 │   ├── utils                   <- Utility scripts
 │   │   ├── inference_example.py    <- Example of inference with trained model
-│   │   └── template_utils.py       <- Extra features for the template
+│   │   └── utils.py                <- Extra features for the template
 │   │
 │   └── train.py                <- Training pipeline
 │
@@ -742,7 +742,7 @@ You can run DDP on mnist example with 4 GPUs like this:
 python run.py trainer.gpus=4 +trainer.accelerator="ddp"
 ```
 ⚠️ When using DDP you have to be careful how you write your models - learn more [here](https://pytorch-lightning.readthedocs.io/en/latest/advanced/multi_gpu.html).
-<br>
+<br><br>
 
 
 ### Extra Features
@@ -1062,7 +1062,7 @@ some_parameter: ${datamodule: some_param}
 ```
 When later accessing this field, say in your lightning model, it will get automatically resolved based on all resolvers that are registered. Remember not to access this field before datamodule is initialized. **You also need to set resolve to false in print_config() in [run.py](run.py) method or it will throw errors!**
 ```python
-template_utils.print_config(config, resolve=False)
+utils.print_config(config, resolve=False)
 ```
 
 </details>

configs/hparams_search/mnist_optuna.yaml

@@ -1,9 +1,9 @@
 # @package _global_
 
 # example hyperparameter optimization of some experiment with Optuna:
-# python run.py -m hparams_search=mnist_optuna experiment=exp_example_simple
-# python run.py -m hparams_search=mnist_optuna experiment=exp_example_simple hydra.sweeper.n_trials=30
-# python run.py -m hparams_search=mnist_optuna experiment=exp_example_simple logger=wandb
+# python run.py -m hparams_search=mnist_optuna experiment=example_simple
+# python run.py -m hparams_search=mnist_optuna experiment=example_simple hydra.sweeper.n_trials=30
+# python run.py -m hparams_search=mnist_optuna experiment=example_simple logger=wandb
 
 defaults:
     - override /hydra/sweeper: optuna

requirements.txt

@@ -5,14 +5,14 @@ pytorch-lightning>=1.2.10
 torchmetrics
 
 # --------- hydra --------- #
-hydra-core==1.1.0.dev5
+hydra-core==1.1.0.dev6
 hydra-colorlog==1.1.0.dev1
 hydra-optuna-sweeper==1.1.0.dev1
 # hydra-ray-launcher==0.1.2
 # hydra-submitit-launcher==1.1.0
 
 # --------- loggers --------- #
-wandb>=0.10.26
+wandb>=0.10.28
 # neptune-client
 # comet-ml
 # mlflow
@@ -21,6 +21,7 @@ wandb>=0.10.26
 rich
 python-dotenv
 pytest
+sh
 pre-commit
 scikit-learn
 pandas

run.py

@@ -13,19 +13,19 @@ def main(config: DictConfig):
     # Imports should be nested inside @hydra.main to optimize tab completion
     # Read more here: https://github.com/facebookresearch/hydra/issues/934
     from src.train import train
-    from src.utils import template_utils
+    from src.utils import utils
 
     # A couple of optional utilities:
     # - disabling python warnings
     # - easier access to debug mode
     # - forcing debug friendly configuration
     # - forcing multi-gpu friendly configuration
     # You can safely get rid of this line if you don't want those
-    template_utils.extras(config)
+    utils.extras(config)
 
     # Pretty print config using Rich library
     if config.get("print_config"):
-        template_utils.print_config(config, resolve=True)
+        utils.print_config(config, resolve=True)
 
     # Train model
     return train(config)

src/__init__.py

@@ -0,0 +1 @@
+#

src/callbacks/wandb_callbacks.py

@@ -13,6 +13,8 @@
 
 
 def get_wandb_logger(trainer: Trainer) -> WandbLogger:
+    """Safely get Weights&Biases logger from Trainer."""
+
     if isinstance(trainer.logger, WandbLogger):
         return trainer.logger
 

src/datamodules/mnist_datamodule.py

@@ -52,6 +52,10 @@ def __init__(
         self.data_val: Optional[Dataset] = None
         self.data_test: Optional[Dataset] = None
 
+    @property
+    def num_classes(self) -> int:
+        return 10
+
     def prepare_data(self):
         """Download data if needed. This method is called only from a single GPU.
         Do not use it to assign state (self.x = y)."""

src/train.py

@@ -7,9 +7,9 @@
 import hydra
 from omegaconf import DictConfig
 
-from src.utils import template_utils
+from src.utils import utils
 
-log = template_utils.get_logger(__name__)
+log = utils.get_logger(__name__)
 
 
 def train(config: DictConfig) -> Optional[float]:
@@ -59,7 +59,7 @@ def train(config: DictConfig) -> Optional[float]:
 
     # Send some parameters from config to all lightning loggers
     log.info("Logging hyperparameters!")
-    template_utils.log_hyperparameters(
+    utils.log_hyperparameters(
         config=config,
         model=model,
         datamodule=datamodule,
@@ -79,7 +79,7 @@ def train(config: DictConfig) -> Optional[float]:
 
     # Make sure everything closed properly
     log.info("Finalizing!")
-    template_utils.finish(
+    utils.finish(
         config=config,
         model=model,
         datamodule=datamodule,
@@ -91,7 +91,7 @@ def train(config: DictConfig) -> Optional[float]:
     # Print path to best checkpoint
     log.info(f"Best checkpoint path:\n{trainer.checkpoint_callback.best_model_path}")
 
-    # Return metric score for Optuna optimization
+    # Return metric score for hyperparameter optimization
     optimized_metric = config.get("optimized_metric")
     if optimized_metric:
         return trainer.callback_metrics[optimized_metric]

src/utils/template_utils.py → src/utils/utils.py

@@ -42,7 +42,7 @@ def extras(config: DictConfig) -> None:
     # enable adding new keys to config
     OmegaConf.set_struct(config, False)
 
-    # disable python warnings if <config.disable_warnings=True>
+    # disable python warnings if <config.ignore_warnings=True>
     if config.get("ignore_warnings"):
         log.info(f"Disabling python warnings! <config.ignore_warnings=True>")
         warnings.filterwarnings("ignore")
@@ -58,6 +58,8 @@ def extras(config: DictConfig) -> None:
         # Debuggers don't like GPUs or multiprocessing
         if config.trainer.get("gpus"):
             config.trainer.gpus = 0
+        if config.datamodule.get("pin_memory"):
+            config.datamodule.pin_memory = False
         if config.datamodule.get("num_workers"):
             config.datamodule.num_workers = 0
 
@@ -128,7 +130,6 @@ def log_hyperparameters(
     """This method controls which parameters from Hydra config are saved by Lightning loggers.
 
     Additionaly saves:
-        - sizes of train, val, test dataset
         - number of trainable model parameters
     """
 
@@ -143,16 +144,6 @@ def log_hyperparameters(
     if "callbacks" in config:
         hparams["callbacks"] = config["callbacks"]
 
-    # save sizes of each dataset
-    # (requires calling `datamodule.setup()` first to initialize datasets)
-    # datamodule.setup()
-    # if hasattr(datamodule, "data_train") and datamodule.data_train:
-    #     hparams["datamodule/train_size"] = len(datamodule.data_train)
-    # if hasattr(datamodule, "data_val") and datamodule.data_val:
-    #     hparams["datamodule/val_size"] = len(datamodule.data_val)
-    # if hasattr(datamodule, "data_test") and datamodule.data_test:
-    #     hparams["datamodule/test_size"] = len(datamodule.data_test)
-
     # save number of model parameters
     hparams["model/params_total"] = sum(p.numel() for p in model.parameters())
     hparams["model/params_trainable"] = sum(