snakeztc / neuraldialog-laed Goto Github PK

afer running ptb-utt.py. I got error
RuntimeError: cuda runtime error (30) : unknown error at /pytorch/aten/src/THC/THCGeneral.cpp:51

BOS and EOS have the same count in the vocabulary. The counter sort the utterances by frequency. Since BOS and EOS have the same count in the vocabulary, the order is random. Sometimes BOS is 4 in vocabulary. Sometimes EOS is 4 in vocabulary.

Traceback (most recent call last):
File "ptb-utt.py", line 155, in
main(config)
File "ptb-utt.py", line 97, in main
prepare_dirs_loggers(config, os.path.basename(file))
File "C:\Users\hw\NeuralDialog-LAED-master\laed\utils.py", line 105, in prepare_dirs_loggers
json.dump(config.dict, fp, indent=4, sort_keys=True)
File "C:\Users\hw\Anaconda3\envs\py35\lib\json_init_.py", line 179, in dump
fp.write(chunk)
TypeError: a bytes-like object is required, not 'str'

Looking at DailyDialogSkipLoader, It's apparent in this line that metadata like dialog acts and emotion are being appended to the dialogue.

dialog.append(Pack(utt=utt, speaker=t_id%2,
                                   meta={'emotion': raw_emotion[t_id], 'act': raw_act[t_id]}))

I'm using a dataset without such metadata, how should I change the code?
Should I put None/0 or remove the meta parameter completely?

It seems like kl_step is not being used while still presented in dailydialog-utt-skip.py and ptb-utt.py. Maybe it's better to comment/delete it so to not confuse people?