Object of type device is not JSON serializable

Hi,

I have trained my TabNet model with pretrainer + regressor with CUDA. However, upon completing the training, if I attempt to save the model, I get the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_932/2022071721.py in <module>
      1 # saving the TABNET model
----> 2 fitted_tbn.selfsupervised.save_model("tbn_0.zip")

/opt/conda/envs/rapids/lib/python3.8/site-packages/pytorch_tabnet/abstract_model.py in save_model(self, path)
    373         # Save models params
    374         with open(Path(path).joinpath("model_params.json"), "w", encoding="utf8") as f:
--> 375             json.dump(saved_params, f, cls=ComplexEncoder)
    376 
    377         # Save state_dict

/opt/conda/envs/rapids/lib/python3.8/json/__init__.py in dump(obj, fp, skipkeys, ensure_ascii, check_circular, allow_nan, cls, indent, separators, default, sort_keys, **kw)
    177     # could accelerate with writelines in some versions of Python, at
    178     # a debuggability cost
--> 179     for chunk in iterable:
    180         fp.write(chunk)
    181 

/opt/conda/envs/rapids/lib/python3.8/json/encoder.py in _iterencode(o, _current_indent_level)
    429             yield from _iterencode_list(o, _current_indent_level)
    430         elif isinstance(o, dict):
--> 431             yield from _iterencode_dict(o, _current_indent_level)
    432         else:
    433             if markers is not None:

/opt/conda/envs/rapids/lib/python3.8/json/encoder.py in _iterencode_dict(dct, _current_indent_level)
    403                 else:
    404                     chunks = _iterencode(value, _current_indent_level)
--> 405                 yield from chunks
    406         if newline_indent is not None:
    407             _current_indent_level -= 1

/opt/conda/envs/rapids/lib/python3.8/json/encoder.py in _iterencode_dict(dct, _current_indent_level)
    403                 else:
    404                     chunks = _iterencode(value, _current_indent_level)
--> 405                 yield from chunks
    406         if newline_indent is not None:
    407             _current_indent_level -= 1

/opt/conda/envs/rapids/lib/python3.8/json/encoder.py in _iterencode(o, _current_indent_level)
    436                     raise ValueError("Circular reference detected")
    437                 markers[markerid] = o
--> 438             o = _default(o)
    439             yield from _iterencode(o, _current_indent_level)
    440             if markers is not None:

/opt/conda/envs/rapids/lib/python3.8/site-packages/pytorch_tabnet/utils.py in default(self, obj)
    337             return int(obj)
    338         # Let the base class default method raise the TypeError
--> 339         return json.JSONEncoder.default(self, obj)

/opt/conda/envs/rapids/lib/python3.8/json/encoder.py in default(self, o)
    177 
    178         """
--> 179         raise TypeError(f'Object of type {o.__class__.__name__} '
    180                         f'is not JSON serializable')
    181 

TypeError: Object of type device is not JSON serializable

I have the following versions

poetry version: N/A
python version: 3.8.12 Operating System: Linux Kernel version 5.4.0-89-generic CUDA: 11.2

I am using Paperspace Gradient with Quadro P5000 16 GB GPU.

About reproducible code:

I am providing as much context as possible since the training data is yielded by a lengthy pipeline. Please let me know if I am to provide it as well.

Thanks in advance for looking into the issue.

[Optimox]

Great!

[tanweer-mahdi]

Thanks for looking into it @Optimox . However, I have solved the issue. Noticing that the error mentioned that the “device” object is not serializable, I found that I was using “cuda:0” as the device name parameter like following:

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

pretrainer_params["optimizer_fn"] = torch.optim.Adam
pretrainer_params["optimizer_params"] = dict(lr = 2e-2)
pretrainer_params["mask_type"] = "entmax"
pretrainer_params["n_shared"] = 1
pretrainer_params["n_independent"] = 1
pretrainer_params["lambda_sparse"] = 0
pretrainer_params["n_d"] = 128
pretrainer_params["n_a"] = 64
pretrainer_params["n_steps"] = 3
pretrainer_params["gamma"] = 1.5
#pretrainer_params["device_name"] = device
pretrainer_params["scheduler_fn"] = torch.optim.lr_scheduler.StepLR
pretrainer_params["scheduler_params"] = {"step_size":10, # how to use learning rate scheduler
                                         "gamma":0.9}

Interestingly, it was not causing any other trouble other than saving the model. I have removed this key from the parameter dictionary and have been able to save/load the model. Thanks again! Please, close the issue if you wish. Cheers!