What am I doing wrong while training a model

[zxcqirara]

I recorded some samples with calling my bot, then tried to train a model according to the guide but it instantly reached 100%, I have recorded 250 samples of calling it and 1000 noise/silence samples. There are command:

rustpotter-cli train -t small --train-dir ./train --test-dir ./test --test-epochs 10 --epochs 2500 -l 0.017 cinet-trained-small.rpw

And logs:

Start training cinet-trained-small.rpw!
Model type: small.
Labels: ["none", "cinet"].
Training with 1250 records.
Testing with 100 records.
Training on 1650ms of audio.
  10 train loss:  0.45068 test acc: 100.00%
  20 train loss:  0.37763 test acc: 100.00%
  30 train loss:  0.33143 test acc: 100.00%
  40 train loss:  0.29268 test acc: 100.00%
  50 train loss:  0.26013 test acc: 100.00%
  60 train loss:  0.23273 test acc: 100.00%
  70 train loss:  0.20961 test acc: 100.00%
  80 train loss:  0.19005 test acc: 100.00%
  90 train loss:  0.17345 test acc: 100.00%
 100 train loss:  0.15933 test acc: 100.00%
   ...
2450 train loss:  0.05909 test acc: 100.00%
2460 train loss:  0.05909 test acc: 100.00%
2470 train loss:  0.05909 test acc: 100.00%
2480 train loss:  0.05909 test acc: 100.00%
2490 train loss:  0.05909 test acc: 100.00%
2500 train loss:  0.05909 test acc: 100.00%
cinet-trained-small.rpw created!

P.S. Files in train folder were called like

[cinet]1.wav
[cinet]2.wav
...
[cinet]249.wav
[cinet]250.wav
1.wav
2.wav
...
999.wav
1000.wav

And in the test folder like:

test1.wav
test2.wav
...
test99.wav
test100.wav

[Dantali0n]

For @zxcqirara you are using a test folder without labels so it can't determine the accuracy

[zxcqirara]

For @zxcqirara you are using a test folder without labels so it can't determine the accuracy

Omg, really 😅

Now I need to find that source code

[zxcqirara]

@Dantali0n, you meant I needed to call them as [test]1.wav, [test]2.wav, etc.?

[Dantali0n]

@Dantali0n, you meant I needed to call them as [test]1.wav, [test]2.wav, etc.?

No, you need to follow the same structure as the train folder and ensure that no files between the test and train folder are the same, they should be unique sets

[zxcqirara]

@Dantali0n, you meant I needed to call them as [test]1.wav, [test]2.wav, etc.?

No, you need to follow the same structure as the train folder and ensure that no files between the test and train folder are the same, they should be unique sets

Got it, but I completely lost those files and tried the training again but faced a problem that I described in GiviMAD/rustpotter#14 (and yes, I named files in both test and training folders according to the same principle)