Experiment 3

Binary classification with validation set implementation.

Parameters

basic setting parameters

clip_length: 5.0 # [sec]

preprocessing parameters

sample_rate: 44100

hop_length: 512

n_fft: 1024

n_mels: 64

Training parameters

number of audio samples: 220500

learning rate: 0.001

batch size: 30

number of epochs: 20

number of samples: 72

balanced dataset: False

random clip cut: False

Classes

Labels

grime jazz

Class distribution

Dataset	Class	Class ID	Samples
Train	`grime` `jazz`	[0, 1]	60
Validation	`grime` `jazz`	[0, 1]	12

Metrics

Logs

Train Epoch: 20 [60/60 (100%)] Loss: 0.353381 Accuracy: 80.0%
[Class: grime] accuracy: 80.0 %
[Class: jazz ] accuracy: 83.3 %
{'grime': 24, 'jazz': 25}
{'grime': 30, 'jazz': 30}
Training for Epoch: [20/20] Loss: 0.392063 Accuracy: 81.66666666666667%
[[=============================================================================================]]
Valid batch: 20 [6/12 (50%)] Loss: 1.069889 Accuracy: 83.33333333333334%
Valid batch: 20 [12/12 (100%)] Loss: 0.136948 Accuracy: 100.0%
Validation for Epoch: [20/20] Loss: 0.603418 Accuracy: 91.7%
[[=============================================================================================]]
{'grime': 5, 'jazz': 6}
{'grime': 6, 'jazz': 6}
[[Validation]] Accuracy for class: grime is 83.3 %
[[Validation]] Accuracy for class: jazz  is 100.0 %
LR:  [0.001]
TRAINING LOSS:  0.39206255972385406
VALIDATION LOSS:  0.6034181341528893
Saving checkpoint: /Users/inhalt/Documents/Gendy/apps/MLVibeCaption/mlvcaudio/models/outputs/checkpoint-epoch20.pth ...
Training is done!
Model trained and stored!