Refactor model hyperparameters

Author: seofernando25

models/cnn.py

@@ -53,15 +53,15 @@ def __init__(
             nn.BatchNorm2d(32),
             nn.ReLU(inplace=True),
             nn.Conv2d(
-                32, 64, kernel_size=min(kernel_size, 3), stride=stride, padding=1
+                32, 64, kernel_size=kernel_size, stride=stride, padding=1
             ),
             nn.BatchNorm2d(64),
             nn.ReLU(inplace=True),
             nn.Conv2d(
                 64,
                 128,
-                kernel_size=min(kernel_size, 3),
-                stride=max(1, stride - 1),
+                kernel_size=kernel_size,
+                stride=stride,
                 padding=1,
             ),
             nn.BatchNorm2d(128),
@@ -267,12 +267,12 @@ def get_param_space(self) -> Dict[str, ParamSpace]:
             "learning_rate": ParamSpace.float_range(
                 min_val=1e-5, max_val=1e-2, default=3e-4
             ),
-            "batch_size": ParamSpace.integer(min_val=16, max_val=128, default=64),
+            "batch_size": ParamSpace.categorical(choices=[16, 32, 64, 128], default=64),
             "weight_decay": ParamSpace.float_range(
                 min_val=0.0, max_val=0.01, default=1e-3
             ),
             "optimizer": ParamSpace.categorical(
-                choices=["AdamW", "Adam", "SGD"], default="AdamW"
+                choices=["AdamW", "SGD"], default="AdamW"
             ),
         }
 
@@ -286,12 +286,6 @@ def _build_optimizer(
                     lr=config.learning_rate,
                     weight_decay=config.weight_decay,
                 )
-            case "Adam":
-                return optim.Adam(
-                    network.parameters(),
-                    lr=config.learning_rate,
-                    weight_decay=config.weight_decay,
-                )
             case "SGD":
                 return optim.SGD(
                     network.parameters(),

models/decision_tree.py

@@ -80,8 +80,5 @@ def get_param_space(self) -> Dict[str, ParamSpace]:
             "min_samples_leaf": ParamSpace.integer(min_val=1, max_val=10, default=2),
             "criterion": ParamSpace.categorical(
                 choices=["gini", "entropy"], default="gini"
-            ),
-            "splitter": ParamSpace.categorical(
-                choices=["best", "random"], default="best"
-            ),
+            )
         }

models/knn.py

@@ -68,13 +68,7 @@ def get_param_space(self) -> Dict[str, ParamSpace]:
             "weights": ParamSpace.categorical(
                 choices=["uniform", "distance"], default="uniform"
             ),
-            "algorithm": ParamSpace.categorical(
-                choices=["auto", "ball_tree", "kd_tree", "brute"], default="auto"
-            ),
-            "p": ParamSpace.categorical(
-                choices=[1, 2], default=2
-            ),  # Manhattan and Euclidean distance
             "metric": ParamSpace.categorical(
-                choices=["minkowski", "chebyshev", "manhattan"], default="minkowski"
+                choices=["minkowski", "manhattan"], default="minkowski"
             ),
         }