Csv logger for spkv (#29)

README.md

@@ -41,5 +41,5 @@ python run.py lightning_datamodule=stp lightning_module=wav2vec2_for_stp lightni
 
 - **Test** [ECAPA2](https://huggingface.co/Jenthe/ECAPA2) for Speaker Verification
 ```
-python run.py lightning_datamodule=spkv lightning_module=ecapa2 ++trainer.limit_train_batches=0
+python run.py lightning_datamodule=spkv lightning_module=ecapa2 logging=csv ++trainer.limit_train_batches=0 ++trainer.limit_val_batches=0
 ```

configs/logging/csv.yaml

@@ -0,0 +1,5 @@
+log_every_n_steps: 100
+
+logger:
+  _target_: lightning.pytorch.loggers.csv_logs.CSVLogger
+  save_dir: "csv/"

run.py

@@ -16,7 +16,8 @@
     seed_everything,
     Trainer,
 )
-from lightning.pytorch.loggers.tensorboard import TensorBoardLogger
+
+from lightning.pytorch.loggers import Logger
 from omegaconf import DictConfig
 
 from torchmetrics import MetricCollection
@@ -51,7 +52,7 @@ def main(cfg: DictConfig):
 
     # Instantiate Trainer
     callbacks: List[Callback] = list(hydra.utils.instantiate(cfg.callbacks).values())
-    logger: TensorBoardLogger = hydra.utils.instantiate(cfg.logging.logger)
+    logger: Logger = hydra.utils.instantiate(cfg.logging.logger)
     trainer: Trainer = hydra.utils.instantiate(
         cfg.trainer, callbacks=callbacks, logger=logger, _convert_="partial"
     )

vibravox/lightning_modules/ecapa2.py

@@ -128,9 +128,6 @@ def on_test_start(self) -> None:
         # Check DataModule parameters
         self.check_datamodule_parameter()
 
-        # Log description
-        self.logger.experiment.add_text(tag="description", text_string=self.description)
-
     def on_test_batch_end(
         self,
         outputs: Dict[str, torch.Tensor],
@@ -188,12 +185,12 @@ def on_test_epoch_end(self):
         Called at the end of the test epoch.
 
         - Triggers the computation of the metrics.
-        - Logs the metrics to tensorboard.
+        - Logs the metrics to the logger (preference: csv for extracting results as there is no training curves to log).
         """
         # Get the metrics as a dict
         metrics_to_log = self.metrics.compute()
 
-        # Log in tensorboard
+        # Log in the logger
         self.log_dict(dictionary=metrics_to_log, sync_dist=True, prog_bar=True)
 
     def check_datamodule_parameter(self) -> None: