Error in training example

[robmarkcole]

Following the example from the README:

ValueError                                Traceback (most recent call last)
<ipython-input-11-f854c515c2ab> in <module>()
----> 1 trainer.fit(model, datamodule=dm)

23 frames
/usr/local/lib/python3.7/dist-packages/torchmetrics/functional/classification/stat_scores.py in _stat_scores_update(preds, target, reduce, mdmc_reduce, num_classes, top_k, threshold, multiclass, ignore_index)
    123         if not mdmc_reduce:
    124             raise ValueError(
--> 125                 "When your inputs are multi-dimensional multi-class, you have to set the `mdmc_reduce` parameter"
    126             )
    127         if mdmc_reduce == "global":

ValueError: When your inputs are multi-dimensional multi-class, you have to set the `mdmc_reduce` parameter

[isaaccorley]

That's odd. I made an example recently that uses the same setup under examples/levircd+.ipynb and didn't run into that error.

When I get a chance I'll try running it locally.

[robmarkcole]

I launched the examples/levircd+.ipynb straight to colab and hit exactly the same issue there.!

1.4 M     Trainable params
0         Non-trainable params
1.4 M     Total params
5.402     Total estimated model params size (MB)
Validation sanity check: 0%
0/2 [00:00<?, ?it/s]
/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/data_loading.py:106: UserWarning: The dataloader, val dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 4 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
  f"The dataloader, {name}, does not have many workers which may be a bottleneck."
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-17-f854c515c2ab> in <module>()
----> 1 trainer.fit(model, datamodule=dm)

23 frames
/usr/local/lib/python3.7/dist-packages/torchmetrics/functional/classification/stat_scores.py in _stat_scores_update(preds, target, reduce, mdmc_reduce, num_classes, top_k, threshold, multiclass, ignore_index)
    123         if not mdmc_reduce:
    124             raise ValueError(
--> 125                 "When your inputs are multi-dimensional multi-class, you have to set the `mdmc_reduce` parameter"
    126             )
    127         if mdmc_reduce == "global":

ValueError: When your inputs are multi-dimensional multi-class, you have to set the `mdmc_reduce` parameter

However the good news is I dont have any issue with examples/probav.ipynb

[isaaccorley]

Looks like the latest torchmetrics==0.5.0 version required a new parameter for computing Accuracy, Precision, Recall on 2D segmentation mask output. Just had to add mdmc_average="global" to each of the metrics in the lightning modules. I'm able to train now so seems like it's fixed. See #25