I’m trying to figure out how to log a wandb confusion matrix as described here inside Lightning. I’m calling self.log inside of my lightning module’s validation_step and logging a custom metric which returns a wandb CustomChart from its compute method and getting the following error:
File "/home/ryan/anaconda3/envs/ml/lib/python3.9/site-packages/pytorch_lightning/core/step_result.py", line 500, in reduce_on_epoch_end
result[k] = torch.tensor(result[k]).float()
RuntimeError: Could not infer dtype of CustomChart
This is one of several ways of tried to get this working and I always get the same error. I can’t seem to find any clue as to what’s going on in Lightning’s logging docs. Anyone know what’s going on? Or at least how to properly log something like this.