Add wandb log

.gitignore

@@ -133,6 +133,7 @@ results/
 ckp/
 checkpoints/
 *.swp
+wandb/
 
 Dockerfile
 build_dgx.sh

micromind/core.py

@@ -25,11 +25,14 @@
 
 # This is used ONLY if you are not using argparse to get the hparams
 default_cfg = {
+    "project_name": "micromind",
     "output_folder": "results",
     "experiment_name": "micromind_exp",
     "opt": "adam",  # this is ignored if you are overriding the configure_optimizers
     "lr": 0.001,  # this is ignored if you are overriding the configure_optimizers
     "debug": False,
+    "log_wandb": False,
+    "wandb_resume": "auto",  # ["allow", "must", "never", "auto" or None]
 }
 
 
@@ -381,14 +384,26 @@ def compute_macs(self, input_shape: Union[List, Tuple]):
 
     def on_train_start(self):
         """Initializes the optimizer, modules and puts the networks on the right
-        devices. Optionally loads checkpoint if already present.
+        devices. Optionally loads checkpoint if already present. It also start wandb
+        logger if selected.
 
         This function gets executed at the beginning of every training.
         """
 
         # pass debug status to checkpointer
         self.checkpointer.debug = self.hparams.debug
 
+        if self.hparams.log_wandb:
+            import wandb
+
+            self.wlog = wandb.init(
+                project=self.hparams.project_name,
+                name=self.hparams.experiment_name,
+                resume=self.hparams.wandb_resume,
+                id=self.hparams.experiment_name,
+                config=self.hparams,
+            )
+
         init_opt = self.configure_optimizers()
         if isinstance(init_opt, list) or isinstance(init_opt, tuple):
             self.opt, self.lr_sched = init_opt
@@ -449,6 +464,8 @@ def init_devices(self):
 
     def on_train_end(self):
         """Runs at the end of each training. Cleans up before exiting."""
+        if self.hparams.log_wandb:
+            self.wlog.finish()
         pass
 
     def eval(self):
@@ -531,6 +548,9 @@ def train(
                     # ok for cos_lr
                     self.lr_sched.step()
 
+                    if self.hparams.log_wandb:
+                        self.wlog.log({"lr": self.lr_sched.get_last_lr()})
+
                 for m in self.metrics:
                     if (
                         self.current_epoch + 1
@@ -574,6 +594,10 @@ def train(
             else:
                 val_metrics = train_metrics.update({"val_loss": loss_epoch / (idx + 1)})
 
+            if self.hparams.log_wandb:  # wandb log
+                self.wlog.log(train_metrics)
+                self.wlog.log(val_metrics)
+
             if e >= 1 and self.debug:
                 break