Empty metrics from precise-test

[ljj7975]

I have correctly set up the dataset but I am seeing 0s for the counts

what might be the cause of this?

Command : precise-test -t 0.85 trained_model.net dataset

Data: <TrainData wake_words=502 not_wake_words=24641 test_wake_words=54 test_not_wake_words=2504>
=== False Positives ===


=== False Negatives ===


=== Counts ===
False Positives: 0
True Negatives: 0
False Negatives: 0
True Positives: 0


=== Summary ===
0 out of 0
0.00%

0.00% false positives
0.00% false negatives

[MatthewScholefield]

Hmm, strange. I don't think this is the case, but can you check if the test files are the correct format? They should be 16 bit integer mono wav files.

[ljj7975]

yea they are .wav with monochannel 16000 sample rates I generated them with librosa write file and for some datasets I generated, it works fine. It just randomly happens that I couldn't figure out why so I was asking here.

As you see in the log, it correctly locate the files Data: <TrainData wake_words=502 not_wake_words=24641 test_wake_words=54 test_not_wake_words=2504>

[ljj7975]

Things still fail with the correct format

import soundfile as sf

data = (item.audio_data.numpy() * 32767).astype('int16')
sf.write(output_path, data, 16000, subtype='PCM_16', endian='LITTLE')

[ljj7975]

I have debugged the issue. pyache was getting into the way.

I think others fall into the same trap. is it possible for u to add support for disabling the cache?

in my cases files might have the same name but contents can be different

[MatthewScholefield]

Ahh, makes sense. It shouldn't be too hard to disable the cache. Out of curiosity, are you working off the dev branch or off of the fork with tf2 support?