# Architecture Multiclass Classifier
## Tests with bird_engine_wind_600 database
- **Parameters**
dataset : ../../databaseAI/images/bird_engine_wind_600
class attribution : ['bird', 'vehicule', 'wind']
learning rate : 0.0001
gradient momentum : 0.9
optimizer : sgd
weight decay : 0.0001
max epochs : 300
accuracy threshold : 70
overfitting tolerance on last epochs : 5
stagnate tolerance on last epochs : 15
### Model multiclass configuration A (Base)
- **Architecture**
receptive_fieds = [3, 8, 8, 16]
- **Results**
Epoch : 200, Training loss : 0.846, Validation loss : 0.816, Accuracy : 63.8 %
[ Stagnation detected ]
=====================================
### Model multiclass configuration B (COLAB)
- **Architecture**
receptive_fields = [3, 8, 8, 8, 8, 8]
- **Results** (Trained on COLAB)
Epoch : 9, Training loss : 1.111, Validation loss : 1.127, Accuracy : 29.95 %
[ Stagnation detected ]
### Model multiclass configuration C (COLAB)
- Architecture
receptive_fields = [3, 16, 16, 16]
- **Results** (Trained on COLAB)
Epoch : 53, Training loss : 0.894, Validation loss : 0.92, Accuracy : 58.07 %
[ Stagnation detected ]
### Model multiclass configuration D
- **Architecture**
receptive_fields = [3, 8, 16, 24]
- **Results**
Epoch : 300, Training loss : 1.036, Validation loss : 1.05, Accuracy : 46.09 %
Training time : 7846.811 seconds
### Model multiclass configuration E
- **Architecture**
receptive_fields = [3, 16, 8, 8]
- **Results**
Epoch : 33, Training loss : 1.091, Validation loss : 1.091, Accuracy : 42.71 %
MANUAL INTERRUPT (STAGNATE)
## Results Table 3 class
| Config | Arch | Result | TERMINATION | Parms |
| ------ | ------------------ | ------ | ----------- | --- |
| B | [3, 8, 8, 8, 8, 8] | 30 % | STAGNATE | |
| D | [3, 8, 16, 24] | 46 % | MAX EPOCHS | |
| C | [3, 16, 16, 16] | 58 % | STAGNATE | |
| A | [3, 8, 8, 16] | 64 % | STAGNATE | |
| E | [3, 16, 8, 8] | 45 % | MANUAL | |
| F | [3, 16, 24, 16, 8] | ? % | | |
## Conclude
Model seem to stagnate, meaning model complexity is enough to learn pattern, but database doesn't allow model to learn more.
***NEED MORE DATA***
# ============================
Many trials with 3 class larger database didn't provide any results, switching to 2 class for validate dataset
# ============================
## Tests with bird_human_970 database
### 2023-05-14 02:53:05.078205 - bird_human_970_params_v1_arch_v2
- **Parameters**
dataset : ../../databaseAI/images/bird_human_970
class attribution : ['bird', 'human']
learning rate : 0.001
gradient momentum : 0.94
optimizer : sgd
weight decay : 1e-05
max epochs : 500
accuracy threshold : 70
overfitting tolerance on last epochs : 15
stagnate tolerance on last epochs : 25
- **Architecture**
receptive_fields = [3, 8, 8, 8, 8, 8]
- **Results**
Epoch : 23
Accuracy : 46.63461538461539 %
Training time : 0:12:00.599619 seconds
### 2023-05-14 02:53:09.476311 - bird_human_970_params_v1_arch_v3
- **Parameters**
No change
- **Architecture**
receptive_fields = [3, 16, 16, 16]
- **Results**
Epoch : 23
Accuracy : 48.55769230769231 %
Training time : 0:14:13.218823 seconds
### 2023-05-14 01:01:44.272442 - bird_human_970_params_v1_arch_v4
- **Parameters**
No change
- **Architecture**
receptive_fields = [3, 8, 16, 24]
- **Results**
Epoch : 63
Accuracy : 49.51923076923077 %
Training time : 0:30:09.139139 seconds
### 2023-05-14 01:31:55.237347 - bird_human_970_params_v1_arch_v4
- **Parameters**
learning rate : 0.0001
- **Architecture**
No change
- **Results**
Epoch : 78
Accuracy : 61.29807692307692 %
Training time : 0:36:28.495559 seconds
## Conclude
Model seem to stagnate, tests with other 2 class database not including human sounds (bird_engine_970) seem to be working well, removing human sounds from db.
***CHANGE DB***
# ============================
## Tests with bird_engine_chainsaw_970 database
================================================================================
### 2023-05-17 02:48:42.915229 - bird_engine_chainsaw_970_default_thin_24
- **Parameters**
dataset : ../../databaseAI/images/bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 85
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 20
- **Architecture**
receptive_fields = [3, 24, 24, 12, 12]
- **Results**
Epoch : 26
Accuracy : 85.36184210526316 %
Training time : 0:38:14.202481 seconds
================================================================================
### 2023-05-17 00:56:10.685347 - bird_engine_chainsaw_970_default_less_pool
- **Parameters**
dataset : ../../databaseAI/images/bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 85
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 20
- **Architecture**
receptive_fields = [3, 32, 32, 16, 16]
- **Results**
Epoch : 45
Accuracy : 85.47149122807018 %
Training time : 0:48:20.269194 seconds
================================================================================
### 2023-05-17 01:44:30.976560 - bird_engine_chainsaw_970_default_fat_32
- **Parameters**
dataset : ../../databaseAI/images/bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 85
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 20
- **Architecture**
receptive_fields = [3, 32, 32, 32, 32]
- **Results**
Epoch : 37
Accuracy : 86.1842105263158 %
Training time : 1:04:11.925233 seconds
## Tests with bird_engine_chainsaw_970 database
================================================================================
### 2023-05-19 17:22:09.993466 - bird_engine_chainsaw_970_leo_less_pool
- **Parameters**
dataset : ../../databaseAI/images/bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : adam
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 32, 32, 16, 16]
- **Results**
Epoch : 49
Accuracy : 87.17105263157895 %
Training time : 0:51:01.788529 seconds
================================================================================
### 2023-05-16 20:49:22.702684 - bird_engine_chainsaw_970_default_2_2
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 2]
- **Results**
Epoch : 49
Accuracy : 81.74342105263158 %
Training time : 0:51:39.972712 seconds
### 2023-05-16 21:41:02.688330 - bird_engine_chainsaw_970_default_2_3
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 4]
- **Results**
Epoch : 49
Accuracy : 87.00657894736842 %
Training time : 0:51:30.718878 seconds
### 2023-05-16 22:32:33.423578 - bird_engine_chainsaw_970_default_2_4
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 6]
- **Results**
Epoch : 49
Accuracy : 88.43201754385964 %
Training time : 0:51:36.063948 seconds
### 2023-05-16 23:24:09.500357 - bird_engine_chainsaw_970_default_2_5
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 8]
- **Results**
Epoch : 49
Accuracy : 80.92105263157895 %
Training time : 0:51:32.612040 seconds
### 2023-05-17 00:15:42.126205 - bird_engine_chainsaw_970_default_2_6
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 10]
- **Results**
Epoch : 49
Accuracy : 86.78728070175438 %
Training time : 0:51:31.266450 seconds
### 2023-05-17 01:07:13.406466 - bird_engine_chainsaw_970_default_2_7
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 12]
- **Results**
Epoch : 49
Accuracy : 87.82894736842105 %
Training time : 0:51:40.962897 seconds
### 2023-05-17 01:58:54.380244 - bird_engine_chainsaw_970_default_2_8
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 20]
- **Results**
Epoch : 49
Accuracy : 84.86842105263158 %
Training time : 0:51:42.338960 seconds
### 2023-05-17 02:50:36.730471 - bird_engine_chainsaw_970_default_2_9
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 24]
- **Results**
Epoch : 49
Accuracy : 85.47149122807018 %
Training time : 0:51:33.998725 seconds
### 2023-05-17 03:42:10.742873 - bird_engine_chainsaw_970_default_2_10
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 32]
- **Results**
Epoch : 49
Accuracy : 87.17105263157895 %
Training time : 0:51:35.672881 seconds
### 2023-05-17 04:33:46.429914 - bird_engine_chainsaw_970_default_2_11
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 38]
- **Results**
Epoch : 49
Accuracy : 87.4451754385965 %
Training time : 0:51:39.509117 seconds
### 2023-05-17 05:25:25.952135 - bird_engine_chainsaw_970_default_2_12
- **Parameters**
dataset : ../../bird_engine_chainsaw_970
class attribution : ['bird', 'engine', 'chainsaw']
learning rate : 0.001
gradient momentum : 0.95
optimizer : sgd
weight decay : 0.0001
max epochs : 50
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 20, 30, 38, 42]
- **Results**
Epoch : 49
Accuracy : 84.4298245614035 %
Training time : 0:51:41.777742 seconds
### 2023-05-16 08:21:24.398931 - bird_engine_chainsaw_970_default_2
- **Parameters**
- **Architecture**
receptive_fields = [3, 20, 30, 38, 16]
- **Results**
Epoch : 27
Accuracy : 85.85526315789474 %
Training time : 0:30:22.465546 seconds
================================================================================
### 2023-05-16 15:22:51.583474 - bird_engine_chainsaw_970_default_2
- **Parameters**
accuracy threshold : 99
- **Architecture**
- **Results**
Epoch : 49
Accuracy : 87.06140350877193 %
Training time : 0:51:41.585158 seconds
### 2023-05-16 08:51:46.887957 - bird_engine_chainsaw_970_default_3
- **Parameters**
- **Architecture**
receptive_fields = [3, 38, 30, 10, 16]
- **Results**
Epoch : 34
Accuracy : 85.47149122807018 %
Training time : 0:52:13.358785 seconds
================================================================================
### 2023-05-16 16:14:33.184635 - bird_engine_chainsaw_970_default_3
- **Parameters**
accuracy threshold : 99
- **Architecture**
- **Results**
Epoch : 49
Accuracy : 86.07456140350878 %
Training time : 1:12:48.567429 seconds
## Tests with new database 6 class : (bird, chainsaw, engine, rain, speech, static)
========================================================
### 2023-09-03 23:10:35.268023 - default_128peak_dropout
- **Parameters**
dataset : ../database/images/
class attribution : ['bird', 'chainsaw', 'engine', 'rain', 'speech', 'static']
learning rate : 0.001
gradient momentum : 0.96
optimizer : sgd
weight decay : 0.0001
max epochs : 75
accuracy threshold : 90
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [3, 32, 64, 64, 64, 32]
- **Results**
Epoch : 26
Accuracy : 90.02659574468085 %
Training time : 1:40:56.620812 seconds
### 2023-10-24 13:24:11.547664 - default_tars
- **Parameters**
dataset : ../database/images/
class attribution : ['bird', 'chainsaw', 'engine', 'rain', 'speech', 'static', 'gunshot']
learning rate : 0.001
gradient momentum : 0.96
optimizer : sgd
weight decay : 0.0001
max epochs : 75
accuracy threshold : 96
overfitting tolerance on last epochs : 10
stagnate tolerance on last epochs : 10
- **Architecture**
receptive_fields = [8, 16, 32, 64, 64]
- **Results**
Epoch : 28
Accuracy : 86.3970588235294 %
Training time : 2:28:14.668570 seconds
Google Drive
Gist
Clipboard
Sign in with Wallet
