Models: [textsum]:TypeError: 'str' does not support the buffer interface

[18811376770]

when i run training command,
$ bazel-bin/textsum/seq2seq_attention \
--mode=train \
--article_key=article \
--abstract_key=abstract \
--data_path=data/training-* \
--vocab_path=data/vocab \
--log_root=textsum/log_root \
--train_dir=textsum/log_root/train
the shell alerts me on data.py file,
Exception in thread Thread-2:
Traceback (most recent call last):
File "/home/led/anaconda2/envs/tensorflow/lib/python3.4/threading.py", line 911, in _bootstrap_inner
self.run()
File "/home/led/anaconda2/envs/tensorflow/lib/python3.4/threading.py", line 859, in run
self._target(self._args, *self._kwargs)
File "/home/led/textsum/textsum/batch_reader.py", line 138, in _FillInputQueue
data.ToSentences(article, include_token=False)]
File "/home/led/textsum/textsum/data.py", line 215, in ToSentences
return [s for s in s_gen]
File "/home/led/textsum/textsum/data.py", line 215, in
return [s for s in s_gen]
File "/home/led/textsum/textsum/data.py", line 189, in SnippetGen
start_p = text.index(start_tok, cur)
TypeError: 'str' does not support the buffer interface
could someone tell me what;s wrong?

[aselle]

That looks like you are using an older not python 3 compatible version of textsum. @nealwu, perhaps can comment. He checked this in:
https://github.com/tensorflow/models/commit/66eb101896a1b7c1691c7416668776ea60ddb78f

Could you please specify what version of textsum you are using i.e. git describe in the root of the models directory.

[nealwu]

That commit was a modification from this pull request: https://github.com/tensorflow/models/pull/1244.

@studynon, did you ever run into an issue like this?

[18811376770]

@aselle
commit d6bee2c713c6aed6522ab32c34b57412d0216d95
Author: Neal Wu [e-mail]
Date: Wed Apr 5 17:22:27 2017 -0700

[nealwu]

Perhaps the model is still not fully compatible with Python 3. It seems like you need to convert the string into bytes in this case. @aselle, do you know how to do that while maintaining Python 2 compatibility?

[aselle]

@nealwu, I've only dabbled with python2 vs python3, but as far as I know the usual approach is to use six functionality as much as possible instead of the python functionality directly. Maybe this
http://six.readthedocs.io/#binary-and-text-data
?

[nealwu]

@18811376770 could you try using six.b() to convert the string to bytes? If that fixes your problem, please submit a pull request for it and we'll merge.

[catfalls]

The following changes help me to run the code in python3. Basically it calls the str.decode() and str.encode() functions in python3:

At around line 189 in data.py:

      start_p = text.index(start_tok.encode(), cur)
      # start_p = text.index(start_tok, cur)
      end_p = text.index(end_tok.encode(), start_p + 1)

At around line 194 in batch_reader.py:

      element = ModelInput(enc_inputs, dec_inputs, targets, enc_input_len,
                           dec_output_len, ' '.join([sent.decode() for sent in abstract_sentences]),
                           ' '.join([sent.decode() for sent in abstract_sentences]))

[skpenn]

The above changes don't work for me. It brings new bugs to the model. My solution is

At around line 189 in data.py:

      #start_p = text.index(start_tok, cur)
      start_p = text.index(start_tok.encode(), cur)
      #end_p = text.index(end_tok, start_p + 1)
      end_p = text.index(end_tok.encode(), start_p + 1)

At last line in data.py:

      #return [s for s in s_gen]
      return [s.decode() for s in s_gen]

[YiHuang2015]

Could I know how can I fix this issue?

[nealwu]

@YiHuang2015 I would suggest asking on StackOverflow.

[YiHuang2015]

Thanks...I will

[karmel]

This question is better asked on StackOverflow since it is not a bug or feature request. There is also a larger community that reads questions there. Thanks!