Open cnewton opened 5 years ago
There's an assumption in params.py
that data_path
will always contain a directory name:
self.data_dir_name = os.path.dirname(data_path)
If data_path is like I have above (cmudict.dict
), then dirname will of course return empty string. Giving a path works fine:
$ g2p-seq2seq --train ~/g2p-seq2seq/cmudict.dict --model_dir g2p-model-nostress --cleanup --reinit
Can this be fixed or the setup directions be clarified to avoid the confusion? Right now README.md
states the training example without a path:
g2p-seq2seq --train train_dictionary.dic --model_dir model_folder_path
I am getting a
WARNING:tensorflow:Estimator's model_fn ... includes params argument, but params are not passed to Estimator.
warning followed by aassert data_dir
(from https://github.com/cmusphinx/g2p-seq2seq/blob/master/g2p_seq2seq/g2p_problem.py#L238) failure when training with the following command (on currentmaster
branch):Package versions used (also reproducible with latest tensorflow / tensor2tensor):
tensor2tensor
==1.6.6
tensorflow
==1.8.0
tensorflow-gpu
==1.8.0
Error log below: