search

shin04 / contrastive-learning-for-audio

2 stars 0 forks source link

dawn stream task [ESC-50] #26

Closed shin04 closed 3 years ago

shin04 commented 3 years ago

result of ESC-50

TEMPLATE

# title

TIMESTAMP: 

## pretrained model

## result

## parameters

## fold
shin04 commented 3 years ago

baseline result

TIMESTAMP: 20210706085113

result

test accuracy: 0.375

validation accuracy
fold1: 0.382500
fold2: 0.377500
fold3: 0.367500
fold4: 0.377500
fold5: 0.410000

parameters

device: cuda
use_pretrained: False
data_format: raw
num_worker: 8
n_epoch: 30
batch_size: 16
lr: 0.001
sr 32000
crop secconds 3
n_mels: 80
frequency shift size: 20

fold

flod 0: {'train': [2, 3, 4], 'valid': [5], 'test': [1]}
flod 1: {'train': [3, 4, 5], 'valid': [1], 'test': [2]}
flod 2: {'train': [1, 4, 5], 'valid': [2], 'test': [3]}
flod 3: {'train': [1, 2, 5], 'valid': [3], 'test': [4]}
flod 4: {'train': [1, 2, 3], 'valid': [4], 'test': [5]}
shin04 commented 3 years ago

pretrained 20210706

postscript I was using incorrect weights!!!

TIMESTAMP: 20210706120141

pretrained model

TIMESTAMP: 20210704200524
computer: mrcomputer02

result

test accuracy: 0.118

validation accuracy
fold1: 0.110000
fold2: 0.102500
fold3: 0.115000
fold4: 0.127500
fold5: 0.155000

parameters

PARAMETERS
device: cuda
use_pretrained: True
data_format: raw
num_worker: 8
n_epoch: 30
batch_size: 32
lr: 0.001
sr 32000
crop secconds 3
n_mels: 80
frequency shift size: 20

fold

FOLD
flod 0: {'train': [2, 3, 4], 'valid': [5], 'test': [1]}
flod 1: {'train': [3, 4, 5], 'valid': [1], 'test': [2]}
flod 2: {'train': [1, 4, 5], 'valid': [2], 'test': [3]}
flod 3: {'train': [1, 2, 5], 'valid': [3], 'test': [4]}
flod 4: {'train': [1, 2, 3], 'valid': [4], 'test': [5]}
shin04 commented 3 years ago

pretrained 20210706 (epoch 100)

postscript I was using incorrect weights!!!

TIMESTAMP: 20210706124942

pretrained model

TIMESTAMP: 20210704200524
computer: mrcomputer02

result

test accuracy: 0.1745

fold 1: 0.180000
fold 2: 0.130000
fold 3: 0.155000
fold 4: 0.165000
fold 5: 0.200000

parameters

PARAMETERS
device: cuda
use_pretrained: True
data_format: raw
num_worker: 8
n_epoch: 100
batch_size: 32
lr: 0.001
sr 32000
crop secconds 3
n_mels: 80
frequency shift size: 20

fold

flod 0: {'train': [2, 3, 4], 'valid': [5], 'test': [1]}
flod 1: {'train': [3, 4, 5], 'valid': [1], 'test': [2]}
flod 2: {'train': [1, 4, 5], 'valid': [2], 'test': [3]}
flod 3: {'train': [1, 2, 5], 'valid': [3], 'test': [4]}
flod 4: {'train': [1, 2, 3], 'valid': [4], 'test': [5]}
shin04 commented 3 years ago

pretrained 20210706

TIMESTAMP: 20210708025853

pretrained model

TIMESTAMP: 20210704200524
computer: mrcomputer03

result

test accuracy: 0.4935

validation accuracy
fold1: 0.482500
fold2: 0.435000
fold3: 0.482500
fold:4: 0.457500
fold:5: 0.507500

parameters

device: cuda
use_pretrained: True
data_format: raw
num_worker: 8
n_epoch: 30
batch_size: 32
lr: 0.001
sr 32000
crop secconds 3
n_mels: 80
frequency shift size: 20

fold

flod 0: {'train': [2, 3, 4], 'valid': [5], 'test': [1]}
flod 1: {'train': [3, 4, 5], 'valid': [1], 'test': [2]}
flod 2: {'train': [1, 4, 5], 'valid': [2], 'test': [3]}
flod 3: {'train': [1, 2, 5], 'valid': [3], 'test': [4]}
flod 4: {'train': [1, 2, 3], 'valid': [4], 'test': [5]}
shin04 commented 3 years ago

unnecessary