fix metrics and bump autrainer version to 0.5.0

README.md

@@ -21,16 +21,16 @@ model-index:
         metrics:
           - type: accuracy
             name: Accuracy
-            value: 0.902352418996893
+            value: 0.6968486462494452
           - type: f1-micro
             name: Micro F1
-            value: 0.902352418996893
+            value: 0.8765212229148116
           - type: f1-macro
             name: Macro F1
-            value: 0.8717470319118755
+            value: 0.8614431334513389
           - type: f1-weighted
             name: Weighted F1
-            value: 0.8999580955196549
+            value: 0.8706722471821455
 ---
 
 # ABGS Ecoacoustic Tagging Model
@@ -65,7 +65,7 @@ The model has been further trained (finetuned) on the training set of the EDANSA
 
 ### Features
 
-The EDANSA2019 dataset was resampled to 32kHz, as this was the sampling rate of AudioSet, where the model was originally trained on. Log mel spectrograms were then extracted using torchlibrosa using the parameters that the upstream model was trained on.
+The EDANSA2019 dataset was resampled to 32kHz, as this was the sampling rate of AudioSet, where the model was originally trained on. Log-Mel spectrograms were then extracted using torchlibrosa using the parameters that the upstream model was trained on.
 
 ### Training process
 
@@ -73,7 +73,7 @@ The model has been trained for 30 epochs. At the end of each epoch, the model wa
 
 ### Evaluation
 
-The model has only been evaluated on in-domain data. The performance on the official test set reached a 0.9 (weighted) f1-score.
+The model has only been evaluated on in-domain data. The performance on the official test set reached a 0.87 (weighted) f1-score.
 
 ## Acknowledgments
 

_best/dev.yaml

@@ -1,27 +1,27 @@
 ml-accuracy:
-  Anth: 0.9706353443673251
-  Bio: 0.9508809396689802
-  Geo: 0.8510411105178858
-  Sil: 0.9583555792845702
-  all: 0.9327282434596903
+  Anth: 0.9679658302189001
+  Bio: 0.9514148424986653
+  Geo: 0.8713294180459157
+  Sil: 0.9615589962626802
+  all: 0.807261078483716
 ml-f1-micro:
-  Anth: 0.9706353443673251
-  Bio: 0.9508809396689802
-  Geo: 0.8510411105178857
-  Sil: 0.9583555792845702
-  all: 0.9327282434596903
+  Anth: 0.9635479951397325
+  Bio: 0.952628839146278
+  Geo: 0.8301620859760395
+  Sil: 0.7049180327868853
+  all: 0.9112810707456978
 ml-f1-macro:
-  Anth: 0.9701027011323622
-  Bio: 0.9508617641355553
-  Geo: 0.8394133823871512
-  Sil: 0.8236346516007533
-  all: 0.8960031248139555
+  Anth: 0.9635479951397325
+  Bio: 0.952628839146278
+  Geo: 0.8301620859760395
+  Sil: 0.7049180327868853
+  all: 0.8628142382622339
 ml-f1-weighted:
-  Anth: 0.9706119080649868
-  Bio: 0.9508871587609018
-  Geo: 0.8468652795820465
-  Sil: 0.9547344914299045
-  all: 0.93077470945946
+  Anth: 0.9635479951397325
+  Bio: 0.952628839146278
+  Geo: 0.8301620859760395
+  Sil: 0.7049180327868853
+  all: 0.9078199657149384
 dev_loss:
   all: 0.18468653024007967
 iteration: 10

_test/test_holistic.yaml

@@ -1,26 +1,26 @@
 ml-accuracy:
-  Anth: 0.9853528628495339
-  Bio: 0.9427430093209055
-  Geo: 0.7696404793608522
-  Sil: 0.9116733244562805
-  all: 0.902352418996893
+  Anth: 0.9849090102086108
+  Bio: 0.9409675987572126
+  Geo: 0.7749667110519307
+  Sil: 0.9298712827341322
+  all: 0.6968486462494452
 ml-f1-micro:
-  Anth: 0.9853528628495338
-  Bio: 0.9427430093209055
-  Geo: 0.7696404793608522
-  Sil: 0.9116733244562805
-  all: 0.902352418996893
+  Anth: 0.9621380846325167
+  Bio: 0.9547465124191903
+  Geo: 0.7757629367536488
+  Sil: 0.7531249999999999
+  all: 0.8765212229148116
 ml-f1-macro:
-  Anth: 0.9769561143673973
-  Bio: 0.937078388485209
-  Geo: 0.7695603976552177
-  Sil: 0.803393227139678
-  all: 0.8717470319118755
+  Anth: 0.9621380846325167
+  Bio: 0.9547465124191903
+  Geo: 0.7757629367536488
+  Sil: 0.7531249999999999
+  all: 0.8614431334513389
 ml-f1-weighted:
-  Anth: 0.9854084295674305
-  Bio: 0.9426173150715926
-  Geo: 0.769005545837607
-  Sil: 0.9028010916019895
-  all: 0.8999580955196549
+  Anth: 0.9621380846325167
+  Bio: 0.9547465124191903
+  Geo: 0.7757629367536488
+  Sil: 0.7531249999999999
+  all: 0.8706722471821455
 loss:
   all: 0.29107842086420826

file_handler.yaml

@@ -1 +1 @@
-$autrainer.datasets.utils.file_handlers.NumpyFileHandler==0.4.0: {}
+$autrainer.datasets.utils.file_handlers.NumpyFileHandler==0.5.0: {}

inference_transform.yaml

@@ -1,2 +1,2 @@
-$autrainer.transforms.smart_compose.SmartCompose==0.4.0:
+$autrainer.transforms.smart_compose.SmartCompose==0.5.0:
   transforms: []

model.yaml

@@ -1,4 +1,4 @@
-$autrainer.models.cnn_10.Cnn10==0.4.0:
+$autrainer.models.cnn_10.Cnn10==0.5.0:
   output_dim: 4
   segmentwise: false
   in_channels: 1

preprocess_file_handler.yaml

@@ -1,2 +1,2 @@
-$autrainer.datasets.utils.file_handlers.AudioFileHandler==0.4.0:
+$autrainer.datasets.utils.file_handlers.AudioFileHandler==0.5.0:
   target_sample_rate: 32000

preprocess_pipeline.yaml

@@ -1,8 +1,8 @@
-$autrainer.transforms.smart_compose.SmartCompose==0.4.0:
+$autrainer.transforms.smart_compose.SmartCompose==0.5.0:
   transforms:
-    - $autrainer.transforms.specific_transforms.StereoToMono==0.4.0:
+    - $autrainer.transforms.specific_transforms.StereoToMono==0.5.0:
         order: -95
-    - $autrainer.transforms.specific_transforms.PannMel==0.4.0:
+    - $autrainer.transforms.specific_transforms.PannMel==0.5.0:
         window_size: 1024
         hop_size: 320
         sample_rate: 32000

target_transform.yaml

@@ -1,4 +1,4 @@
-$autrainer.datasets.utils.target_transforms.MultiLabelEncoder==0.4.0:
+$autrainer.datasets.utils.target_transforms.MultiLabelEncoder==0.5.0:
   threshold: 0.5
   labels:
     - Anth