Synchronize with DENGE

configs/boston_dataset.yaml

@@ -0,0 +1,61 @@
+estimator: # basic MLP layer configuration
+  class: 'ensemble'
+  num_networks : 5
+  network:
+    estimator_network:
+      - fc1 : {class : Linear, in_features : 13, out_features : 50}
+      - projection : {class : LinearVarianceNetworkHead, in_features : 50, out_features : 1}
+    predictor_network: 
+      - fc1 : {class : Linear, in_features : 13, out_features : 50}
+      - projection : {class : Linear, in_features : 50, out_features : 1}
+  optimizer:
+    class : 'Adam'
+    lr : 0.1
+dataset:
+  class: 'xls'
+  xls_path: "regression_datasets/boston.csv"
+  batch_size : 100
+  cv_split_num: 5
+  test_ratio: 0.10
+transforms:
+    train:
+      x :
+        - {
+          class : Standardize,
+          mean : [3.613523557312254, 11.363636363636363, 11.13677865612648, 0.0691699604743083, 0.5546950592885376, 6.284634387351779, 68.57490118577076, 3.795042687747036, 9.549407114624506, 408.2371541501976, 18.455533596837945, 356.6740316205534, 12.653063241106722], 
+          std : [8.60154510533249, 23.32245299451514, 6.860352940897585, 0.25399404134041037, 0.11587767566755595, 0.7026171434153233, 28.148861406903617, 2.105710126627611, 8.707259384239366, 168.53711605495903, 2.1649455237144406, 91.29486438415783, 7.141061511348571]
+          }
+      y :
+        - {class : Standardize, mean : [22.532806324110677], std : [9.197104087379818]}
+    val:
+      x :
+        - {
+          class : Standardize,
+          mean : [3.613523557312254, 11.363636363636363, 11.13677865612648, 0.0691699604743083, 0.5546950592885376, 6.284634387351779, 68.57490118577076, 3.795042687747036, 9.549407114624506, 408.2371541501976, 18.455533596837945, 356.6740316205534, 12.653063241106722], 
+          std : [8.60154510533249, 23.32245299451514, 6.860352940897585, 0.25399404134041037, 0.11587767566755595, 0.7026171434153233, 28.148861406903617, 2.105710126627611, 8.707259384239366, 168.53711605495903, 2.1649455237144406, 91.29486438415783, 7.141061511348571]
+          }
+      y :
+        - {class : Standardize, mean : [22.532806324110677], std : [9.197104087379818]}
+train:
+  train_type : epoch
+  num_iter : 40
+  weight_type : both
+# logger:
+#   type: 'wandb'
+#   project: 'uncertainty-estimation'
+#   entity: 'kbora'
+#   name: 'Boston Dataset (Baseline,MAE)'
+metrics:
+  categorize : True
+  list:
+    train:
+      - "rmse"
+      - "mae"
+    val:
+      - "rmse"
+      - "mae"
+      - "gm"
+    test:
+      - "rmse"
+      - "mae"
+      - "gm"

configs/concrete_dataset.yaml

@@ -0,0 +1,62 @@
+estimator: # basic MLP layer configuration
+  class: 'ensemble'
+  num_networks : 5
+  network:
+    estimator_network:
+      - fc1 : {class : Linear, in_features : 8, out_features : 50}
+      - projection : {class : LinearVarianceNetworkHead, in_features : 50, out_features : 1}
+    predictor_network: 
+      - fc1 : {class : Linear, in_features : 8, out_features : 50}
+      - projection : {class : Linear, in_features : 50, out_features : 1}
+  optimizer:
+    class : 'Adam'
+    lr : 0.01
+dataset:
+  class: 'xls'
+  xls_path: "regression_datasets/Concrete_Data.xls"
+  batch_size : 512
+  cv_split_num: 1
+  test_ratio: 0.10
+transforms:
+    train:
+      x :
+        - {
+          class : Standardize,
+          mean : [281.16563106796116, 73.89548543689321, 54.1871359223301, 181.56635922330096, 6.203111650485437, 972.9185922330097, 773.5788834951456, 45.662135922330094], 
+          std : [104.5071416428718, 86.27910364316895, 63.99646938186508, 21.355567066911522, 5.973491650590111, 77.75381809178927, 80.1754273990239, 63.16991158103249]
+          }
+      y :
+        - {class : Standardize, mean : [35.8178358261136], std : [16.705679174867946]}
+    val:
+      x : 
+        - {
+          class : Standardize,
+          mean : [281.16563106796116, 73.89548543689321, 54.1871359223301, 181.56635922330096, 6.203111650485437, 972.9185922330097, 773.5788834951456, 45.662135922330094], 
+          std : [104.5071416428718, 86.27910364316895, 63.99646938186508, 21.355567066911522, 5.973491650590111, 77.75381809178927, 80.1754273990239, 63.16991158103249]
+          }
+      y : 
+        - {class : Standardize, mean : [35.8178358261136], std : [16.705679174867946]}
+train:
+  train_type : epoch
+  num_iter : 40
+  val_every : 40
+  weight_type : both
+# logger:
+#   type: 'wandb'
+#   project: 'uncertainty-estimation'
+#   entity: 'kbora'
+#   name: 'Concrete Dataset (Baseline,MAE)'
+metrics:
+  categorize : True
+  list:
+    train:
+      - "rmse"
+      - "mae"
+    val:
+      - "rmse"
+      - "mae"
+      - "gm"
+    test:
+      - "rmse"
+      - "mae"
+      - "gm"

configs/imdb_wiki_dataset.yaml

@@ -40,7 +40,7 @@ logger:
   type: 'wandb'
   project: 'uncertainty-estimation'
   entity: 'kbora'
-  name: 'IMDB-WIKI R50 Weighted (Baseline)'
+  name: 'IMDB-WIKI R50 Estimator'
 metrics:
   categorize : True
   list:

configs/imdb_wiki_dataset_aleatoric.yaml

@@ -0,0 +1,57 @@
+estimator: # basic MLP layer configuration
+  class: 'ensemble'
+  num_networks : 5
+  network:
+    estimator_network:
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : LinearVarianceNetworkHead}
+    predictor_network: 
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : Linear}
+  optimizer:
+    class : 'Adam'
+    lr : 0.001
+    scheduler : {class : MultiStepLR, milestones : [60,80], gamma : 0.1}
+dataset:
+  class: 'imdb-wiki'
+  data_dir : "regression_datasets/imdb_wiki/data"
+  batch_size : 64
+transforms:
+    train:
+      x :
+        - {class : Resize, size : [224, 224]}
+        - {class : RandomCrop, size : 224, padding : 16}
+        - {class : RandomHorizontalFlip}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y :
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+    val:
+      x : 
+        - {class : Resize, size : [224, 224]}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y : 
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+train:
+  train_type : epoch
+  num_iter : 90
+  val_every : 1
+  weight_type : 'aleatoric'
+logger:
+  type: 'wandb'
+  project: 'uncertainty-estimation'
+  entity: 'kbora'
+  name: 'IMDB-WIKI R50 Weighted (Aleatoric, MAE)'
+metrics:
+  categorize : True
+  list:
+    train:
+      - "rmse"
+      - "mae"
+    val:
+      - "rmse"
+      - "mae"
+      - "gm"
+    test:
+      - "rmse"
+      - "mae"
+      - "gm"

configs/imdb_wiki_dataset_baseline.yaml

@@ -0,0 +1,57 @@
+estimator: # basic MLP layer configuration
+  class: 'ensemble'
+  num_networks : 5
+  network:
+    estimator_network:
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : LinearVarianceNetworkHead}
+    predictor_network: 
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : Linear}
+  optimizer:
+    class : 'Adam'
+    lr : 0.001
+    scheduler : {class : MultiStepLR, milestones : [60,80], gamma : 0.1}
+dataset:
+  class: 'imdb-wiki'
+  data_dir : "regression_datasets/imdb_wiki/data"
+  batch_size : 64
+transforms:
+    train:
+      x :
+        - {class : Resize, size : [224, 224]}
+        - {class : RandomCrop, size : 224, padding : 16}
+        - {class : RandomHorizontalFlip}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y :
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+    val:
+      x : 
+        - {class : Resize, size : [224, 224]}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y : 
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+train:
+  train_type : epoch
+  num_iter : 90
+  val_every : 1
+  weight_type : 'baseline'
+logger:
+  type: 'wandb'
+  project: 'uncertainty-estimation'
+  entity: 'kbora'
+  name: 'IMDB-WIKI R50 Weighted (Baseline, MAE)'
+metrics:
+  categorize : True
+  list:
+    train:
+      - "rmse"
+      - "mae"
+    val:
+      - "rmse"
+      - "mae"
+      - "gm"
+    test:
+      - "rmse"
+      - "mae"
+      - "gm"

configs/imdb_wiki_dataset_epistemic.yaml

@@ -0,0 +1,57 @@
+estimator: # basic MLP layer configuration
+  class: 'ensemble'
+  num_networks : 5
+  network:
+    estimator_network:
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : LinearVarianceNetworkHead}
+    predictor_network: 
+      - resnet_branch : {class : ResNet, block : Bottleneck, layers : [3,4,6,3], pred_head : Linear}
+  optimizer:
+    class : 'Adam'
+    lr : 0.001
+    scheduler : {class : MultiStepLR, milestones : [60,80], gamma : 0.1}
+dataset:
+  class: 'imdb-wiki'
+  data_dir : "regression_datasets/imdb_wiki/data"
+  batch_size : 64
+transforms:
+    train:
+      x :
+        - {class : Resize, size : [224, 224]}
+        - {class : RandomCrop, size : 224, padding : 16}
+        - {class : RandomHorizontalFlip}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y :
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+    val:
+      x : 
+        - {class : Resize, size : [224, 224]}
+        - {class : ToTensor}
+        - {class : Normalize, mean : [.5, .5, .5], std : [.5, .5, .5]}
+      y : 
+        - {class : MinMaxNormalize, min_val : [0], max_val : [186]}
+train:
+  train_type : epoch
+  num_iter : 90
+  val_every : 1
+  weight_type : 'epistemic'
+logger:
+  type: 'wandb'
+  project: 'uncertainty-estimation'
+  entity: 'kbora'
+  name: 'IMDB-WIKI R50 Weighted (Epistemic, MAE)'
+metrics:
+  categorize : True
+  list:
+    train:
+      - "rmse"
+      - "mae"
+    val:
+      - "rmse"
+      - "mae"
+      - "gm"
+    test:
+      - "rmse"
+      - "mae"
+      - "gm"