Add indication of lowest validation loss epochs on loss plots

This MR closes #70 (closed) by adding vertical line indicators for the epochs in which the validation loss is at its lowest. The vertical lines fade as we go from the lowest, to the second, third and (up-to) fourth lowest values.

src/mednet/engine/callbacks.py

@@ -40,9 +40,9 @@ class LoggingCallback(lightning.pytorch.Callback):
         super().__init__()
 
         # timers
-        self._start_training_time = 0.0
-        self._start_training_epoch_time = 0.0
-        self._start_validation_epoch_time = 0.0
+        self._start_training_time = time.time()
+        self._start_training_epoch_time = time.time()
+        self._start_validation_epoch_time = time.time()
 
         # log accumulators for a single flush at each training cycle
         self._to_log: dict[str, float] = {}
@@ -255,7 +255,7 @@ class LoggingCallback(lightning.pytorch.Callback):
                 f"interval to a suitable value, so it allows some measures "
                 f"to be performed. Note this is only possible if the time "
                 f"to log a single measurement point is smaller than the "
-                f"time it takes to train a single epoch."
+                f"time it takes to **train** a single epoch."
             )
         else:
             for metric_name, metric_value in aggregate(metrics).items():
@@ -308,7 +308,7 @@ class LoggingCallback(lightning.pytorch.Callback):
                 f"interval to a suitable value, so it allows some measures "
                 f"to be performed. Note this is only possible if the time "
                 f"to log a single measurement point is smaller than the "
-                f"time it takes to train a single epoch."
+                f"time it takes to **validate** a single epoch."
             )
         else:
             for metric_name, metric_value in aggregate(metrics).items():