experiments/STT
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Reverse ordered loss again.

Browse Source
This commit is contained in:
Daniel 2020-02-06 12:49:46 +01:00
parent f5145526f0
commit 4186cbef88
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
util/evaluate_tools.py
View File

@ -65,8 +65,9 @@ def calculate_report(wav_filenames, labels, decodings, losses):
# Getting the WER and CER from the accumulated edit distances and lengths # Getting the WER and CER from the accumulated edit distances and lengths
samples_wer, samples_cer = wer_cer_batch(samples) samples_wer, samples_cer = wer_cer_batch(samples)
# Order the remaining items by their loss (lowest loss on top) # Reversed because the worst WER with the best loss is to identify systemic issues, where the acoustic model is confident,
samples.sort(key=lambda s: s.loss) # yet the result is completely off the mark. This can point to transcription errors and stuff like that.
samples.sort(key=lambda s: s.loss, reverse=True)
# Then order by ascending WER/CER # Then order by ascending WER/CER
if FLAGS.utf8: if FLAGS.utf8: