TensorBoard Metadata UnicodeDecodeError

Question

I have a TensorBoard embedding that is created and saved like this from a gensim Doc2Vec model:

embedding = model.docvecs.vectors_docs

tf.reset_default_graph()
sess = tf.InteractiveSession()

X = tf.Variable([0.0], name='embedding')
place = tf.placeholder(tf.float32, shape=embedding.shape)
set_x = tf.assign(X, place, validate_shape=False)

sess.run(tf.global_variables_initializer())
sess.run(set_x, feed_dict={place: embedding})

summary_writer = tf.summary.FileWriter('log', sess.graph)

config = projector.ProjectorConfig()
embedding_conf = config.embeddings.add()
embedding_conf.tensor_name = 'embedding:0'
embedding_conf.metadata_path = os.path.join('log','metadata.tsv')

projector.visualize_embeddings(summary_writer, config)

saver = tf.train.Saver([X])
saver.save(sess, os.path.join('log', 'model.ckpt'), 1)

Results in this error:

TensorBoard 1.5.1 at http://COMPUTER_NAME:6006 (Press CTRL+C to quit)
E0222 17:15:20.231085 Thread-1 _internal.py:88] Error on request:
Traceback (most recent call last):
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\werkzeug
\serving.py", line 270, in run_wsgi
    execute(self.server.app)
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\werkzeug
\serving.py", line 258, in execute
    application_iter = app(environ, start_response)
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorbo
ard\backend\application.py", line 271, in __call__
    return self.data_applications[clean_path](environ, start_response)
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\werkzeug
\wrappers.py", line 308, in application
    resp = f(*args[:-2] + (request,))
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorbo
ard\plugins\projector\projector_plugin.py", line 514, in _serve_metadata
    for line in f:
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\lib\io\file_io.py", line 214, in __next__
    return self.next()
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\lib\io\file_io.py", line 208, in next
    retval = self.readline()
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\lib\io\file_io.py", line 178, in readline
    return self._prepare_value(self._read_buf.ReadLineAsString())
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\lib\io\file_io.py", line 94, in _prepare_value
    return compat.as_str_any(val)
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\util\compat.py", line 106, in as_str_any
    return as_str(value)
  File "c:\users\user\_installed\anaconda\envs\docsim\lib\site-packages\tensorfl
ow\python\util\compat.py", line 84, in as_text
    return bytes_or_text.decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x92 in position 84: invalid
 start byte

When I remove the embedding_conf.metadata_path = os.path.join('log','metadata.tsv') line, no error occurs. In addition, I can then select the very same metadata file that TensorFlow was attempting to bind to the embedding when the error occurred.

Why is this error occurring?

Answer 1

Is it possible that the data inside your metadata.tsv file isn't UTF-8 encoded?