2

I've been playing with SyntaxNet. After each run, a line like this is printed:

INFO:tensorflow:Seconds elapsed in evaluation: 20.65, eval metric: 11.48%

I've traced it to line 134 of parser_eval.py:

tf_eval_epochs, tf_eval_metrics, tf_documents = sess.run([
    parser.evaluation['epochs'],
    parser.evaluation['eval_metrics'],
    parser.evaluation['documents'],
])

num_tokens += tf_eval_metrics[0]
num_correct += tf_eval_metrics[1]
(...)
eval_metric = 100.0 * num_correct / num_tokens
(...)
logging.info('Seconds elapsed in evaluation: %.2f, '
             'eval metric: %.2f%%', time.time() - t, eval_metric)

It seems to be a measure of accuracy, but how is it calculated and what does it mean?

Undo
  • 25,519
  • 37
  • 106
  • 129
  • I'm also looking for info regrading this. – Kahless Sep 10 '16 at 22:29
  • Where you able to figure it out? I am looking to change the number of lines processed at a time (Syntaxnet procceses 1024 at a time). I also traced it until this part. I don't understand how `sess.run` works. – kskp Sep 30 '16 at 19:00
  • @kskp Nope, didn't get any further than the digging in the question. Sess.run is a Tensorflow thing, though. – Undo Sep 30 '16 at 20:34

0 Answers0