ValueError: NaN encountered in calc_diversity when enabling diversity evaluation

[111bnh]

Hi，

I encountered an issue when running evaluate.py. In order to check the diversity metrics of the model, I uncommented lines 227-228 to enable diversity.calc_diversity(): [https://github.com/iSEE-Laboratory/DGTR/blob/226bf457483330aa8d6be464e69c25305428caa0/tools/evaluate.py#L227-L228]

DGTR/tools/evaluate.py

Lines 227 to 228 in 226bf45

	# diversity = DiversityEvalator(temp_result_path, performance_results)
	# final_result = diversity.calc_diversity()

However, after the evaluation progress reached 100%, the program crashed during the final calculation step. The traceback indicates that NaN values were generated in the probs tensor within the _calc_entropy function, which caused torch.distributions.Categorical to fail because the Simplex constraint was not satisfied.

The specific error log is as follows:
Evaluate: 100%|██████████████████████████████████| 1126/1126 [3:59:08<00:00, 12.74s/it, q1=0.1, pen=0.01, valid_q1=0.03]

Finished in 14362.68s
Removing ./Experiments/dgtr/test_results/epoch_25/metrics_0_1126.json

Traceback (most recent call last):
File "./tools/evaluate.py", line 228, in
final_result = diversity.calc_diversity()
File "./tools/evaluate.py", line 127, in calc_diversity
mean_entropy_joints = self._calc_entropy(joint_angles)
File "./tools/evaluate.py", line 181, in _calc_entropy
distribution = torch.distributions.Categorical(probs=prob)
File ".../torch/distributions/categorical.py", line 64, in init
super(Categorical, self).init(batch_shape, validate_args=validate_args)
File ".../torch/distributions/distribution.py", line 55, in init
raise ValueError(
ValueError: Expected parameter probs (Tensor of shape (100,)) of distribution Categorical(probs: torch.Size([100])) to satisfy the constraint Simplex(), but found invalid values:
tensor([nan, nan, ... (omitted) ... nan])

Could you please advise on what might be causing this? Is it likely due to invalid generated joint angles, or is there a potential division-by-zero issue in the entropy calculation logic?

Thanks in advance for your help!

DGTR/tools/evaluate.py

Lines 227 to 228 in 226bf45

	# diversity = DiversityEvalator(temp_result_path, performance_results)
	# final_result = diversity.calc_diversity()