Error while training model with Categorical only data

[deepakchandra30]

Hi guys,

I'm facing an issue while training the model with a dataset which doesn't have any numeric columns. Can you help me fixing this issue?

My info file

    "header": "infer",
    "column_names": null,
    "num_col_idx": [], 
    "cat_col_idx": [0, 1, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14], 
    "target_col_idx": [3],  
    "file_type": "csv",

Error/Output

==============Starting Trainin Loop, total number of epoch = 8000==============
Epoch 1/8000: 100%|█| 8/8 [00:04<00:00,  1.66it/s, lr=0.0005, DLoss=2.52, CLoss=0, TotalLoss=2.52, closs_weight=1, dloss_weight=
Traceback (most recent call last):
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/main.py", line 46, in <module>
    tabdiff_main(args)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/main.py", line 283, in main
    trainer.run_loop()
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 198, in run_loop
    num_noise_dict = {"num_noise/rho": self.diffusion.num_schedule.rho().item()}
RuntimeError: a Tensor with 0 elements cannot be converted to Scalar

[juntongshi48]

Hi deepakchandra30,

Thank you for reporting this!

The issue was caused by our oversight when implementing the logging of scheduler parameters: we didn't account for the case where numerical features were empty. We've just fixed it up. Please check the latest commit 75040bb.

Juntong

[deepakchandra30]

You guys are really quick with replies and solutions. Thanks a lot

Regards,
Deepak

[deepakchandra30]

Hi @juntongshi48 ,

I think the recent commit didn't fix the issue completely, could you help me with this please.

Epoch 496/8000: 100%|█| 8/8 [00:05<00:00,  1.58it/s, lr=0.000405, DLoss=1.63, CLoss=0, TotalLoss=1.63, closs_weight=0.938, dloss_weigh
Epoch 497/8000: 100%|█| 8/8 [00:04<00:00,  1.64it/s, lr=0.000405, DLoss=1.45, CLoss=0, TotalLoss=1.45, closs_weight=0.938, dloss_weigh
Epoch 498/8000: 100%|█| 8/8 [00:04<00:00,  1.64it/s, lr=0.000405, DLoss=1.48, CLoss=0, TotalLoss=1.48, closs_weight=0.938, dloss_weigh
Epoch 499/8000: 100%|█| 8/8 [00:04<00:00,  1.64it/s, lr=0.000405, DLoss=1.52, CLoss=0, TotalLoss=1.52, closs_weight=0.938, dloss_weigh
Epoch 500/8000: 100%|█| 8/8 [00:05<00:00,  1.53it/s, lr=0.000405, DLoss=1.51, CLoss=0, TotalLoss=1.51, closs_weight=0.938, dloss_weigh
==============Routine Generation Evaluation every 500, currently at epoch #500, wiht total_loss=1.5094.==============                
==============Starting Sampling, total samples to generate = 900==============                                                       
Samples left to generate: 900
Traceback (most recent call last):
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/main.py", line 46, in <module>                                      
    tabdiff_main(args)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/main.py", line 283, in main                                 
    trainer.run_loop()
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 276, in run_loop                          
    out_metrics, _, _ = self.evaluate_generation(save_metric_details=True, plot_density=True)                                        
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 406, in evaluate_generation               
    syn_df = self.sample_synthetic(num_samples, ema=ema)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 455, in sample_synthetic                  
    syn_data = self.diffusion.sample_all(num_samples, self.sample_batch_size, keep_nan_samples=keep_nan_samples)                     
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/models/unified_ctime_diffusion.py", line 227, in sample_all 
    sample = self.sample(b)
  File "/home/dnallamothu/synthetic-data-generation/.venv/lib64/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/models/unified_ctime_diffusion.py", line 174, in sample     
    t_hat = torch.min(t_hat, dim=-1, keepdim=True).values    # take the samllest t_hat induced by sigma_num                          
IndexError: min(): Expected reduction dim 1 to have non-zero size.```

[juntongshi48]

should be good now

[deepakchandra30]

Hi @juntongshi48,

Can you help me with this please.
I think there are some more issues,

==============Starting Sampling, total samples to generate = 900==============
Samples left to generate: 900
Sampling Progress: 100%|████████████████████████████████████████████████████████████████████████| 50/50 [01:38<00:00,  1.96s/it]
Shape of the generated sample = torch.Size([900, 15])
Traceback (most recent call last):
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/main.py", line 46, in <module>
    tabdiff_main(args)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/main.py", line 283, in main
    trainer.run_loop()
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 276, in run_loop
    out_metrics, _, _ = self.evaluate_generation(save_metric_details=True, plot_density=True)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 406, in evaluate_generation
    syn_df = self.sample_synthetic(num_samples, ema=ema)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 479, in sample_synthetic
    syn_num, syn_cat, syn_target = split_num_cat_target(syn_data, info, num_inverse, int_inverse, cat_inverse) 
  File "/home/dnallamothu/synthetic-data-generation/.venv/lib64/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/home/dnallamothu/synthetic-data-generation/TabDiff-main/tabdiff/trainer.py", line 588, in split_num_cat_target
    syn_num = num_inverse(syn_num).astype(np.float32)
AttributeError: 'Tensor' object has no attribute 'astype'

[deepakchandra30]

Hi guys, Can someone help me with this? @juntongshi48