From 6d8249ffc1608556bae9f809f95a80230551efbc Mon Sep 17 00:00:00 2001
From: Olga Lyashevska <olga@herenstraat.nl>
Date: Mon, 10 Jul 2023 09:38:09 +0000
Subject: [PATCH 1/6] experimenting 3 layers

---
 bird_cloud_gnn/gnn_model.py | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index e25ab2b..9d02eea 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -40,8 +40,12 @@ def __init__(self, in_feats: int, h_feats: int, num_classes: int):
         self.in_feats = in_feats
         self.h_feats = h_feats
         self.num_classes = num_classes
+        # self.conv1 = GraphConv(in_feats, h_feats)
+        # self.conv2 = GraphConv(h_feats, h_feats)
+        # self.conv3 = GraphConv(h_feats, num_classes)
         self.conv1 = GraphConv(in_feats, h_feats)
         self.conv2 = GraphConv(h_feats, num_classes)
+        # self.dropout = nn.Dropout(0.2)
 
     def oneline_description(self):
         """Description of the model to uniquely identify it in logs"""
@@ -70,6 +74,9 @@ def forward(self, g, in_feat):
         h = self.conv1(g, in_feat)
         h = F.relu(h)
         h = self.conv2(g, h)
+        # h = F.relu(h)
+        # h = self.dropout(h)
+        # h = self.conv3(g, h)
         g.ndata["h"] = h
         return dgl.mean_nodes(g, "h")
 
@@ -222,8 +229,12 @@ def fit_and_evaluate(
             epoch_values["Accuracy/test"] = num_correct / num_total
             epoch_values["Layer/conv1"] = self.conv1.weight.detach()
             epoch_values["Layer/conv2"] = self.conv2.weight.detach()
+<<<<<<< HEAD
             for i, pg in enumerate(optimizer.param_groups):
                 epoch_values[f"LearningRate/ParGrp{i}"] = pg["lr"]
+=======
+            # epoch_values["Layer/conv3"] = self.conv3.weight.detach()
+>>>>>>> experimenting 3 layers
             if self.num_classes == 2:
                 epoch_values["FalseNegativeRate/test"] = num_false_negative / num_total
                 epoch_values["FalsePositiveRate/test"] = num_false_positive / num_total

From 541c5b54db92230e8a3de1c7da11529c4570cfd7 Mon Sep 17 00:00:00 2001
From: Olga Lyashevska <olga@herenstraat.nl>
Date: Tue, 11 Jul 2023 14:12:49 +0000
Subject: [PATCH 2/6] Added dynamic layers

---
 bird_cloud_gnn/gnn_model.py | 63 +++++++++++++++++++------------------
 tests/test_gnn_model.py     | 11 ++++---
 2 files changed, 38 insertions(+), 36 deletions(-)

diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index 9d02eea..efe76da 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -3,12 +3,13 @@
 import os
 import dgl
 import numpy as np
-import torch.nn.functional as F
+
 from dgl.dataloading import GraphDataLoader
 from dgl.nn.pytorch.conv import GraphConv
 from torch import nn
 from torch import optim
 from tqdm import tqdm
+from torch.nn.modules import Module
 
 
 os.environ["DGLBACKEND"] = "pytorch"
@@ -17,11 +18,11 @@
 class GCN(nn.Module):
     """Graph Convolutional Network construction module
 
-    A two-layer GCN is constructed from input dimension, hidden dimensions and number of classes.
-    Each layer computes new node representations by aggregating neighbor information.
+    A n-layer GCN is constructed from input features and list of layers
+    Each layer computes new node representations by aggregating neighbour information.
     """
 
-    def __init__(self, in_feats: int, h_feats: int, num_classes: int):
+    def __init__(self, in_feats: int, layers_data: list):
         """
         The __init__ function is the constructor for a class. It is called when an object of that class is instantiated.
         It can have multiple arguments and it will always be called before __new__().
@@ -30,36 +31,38 @@ def __init__(self, in_feats: int, h_feats: int, num_classes: int):
         Args:
             self: Access variables that belongs to the class object
             in_feats: the number of input features
-            h_feats: the number of hidden features that we want to use for our first graph convolutional layer
-            num_classes: the number of classes that we want to predict
+            layers_data: is a list of tuples of size of hidden layer and activation function
 
         Returns:
             The self object
         """
         super().__init__()
         self.in_feats = in_feats
-        self.h_feats = h_feats
-        self.num_classes = num_classes
-        # self.conv1 = GraphConv(in_feats, h_feats)
-        # self.conv2 = GraphConv(h_feats, h_feats)
-        # self.conv3 = GraphConv(h_feats, num_classes)
-        self.conv1 = GraphConv(in_feats, h_feats)
-        self.conv2 = GraphConv(h_feats, num_classes)
-        # self.dropout = nn.Dropout(0.2)
+        self.layers = nn.ModuleList()
+        self.name = ""
+        for size, activation in layers_data:
+            self.layers.append(GraphConv(in_feats, size))
+            self.name = self.name + f"{in_feats}-{size}_"
+            in_feats = size  # For the next layer
+            if activation is not None:
+                assert isinstance(
+                    activation, Module
+                ), "Each tuples should contain a size (int) and a torch.nn.modules.Module."
+                self.layers.append(activation)
+                self.name = self.name + "ReLU_"
+            self.num_classes = size  # the last size should correspond to the number of classes were predicting
 
     def oneline_description(self):
         """Description of the model to uniquely identify it in logs"""
         return "-".join(
             [
-                "in",
-                f"GC_{self.h_feats}",
-                "RELU",
-                f"GC_{self.num_classes}",
+                "in_",
+                f"{self.name}",
                 "mean-out",
             ]
         )
 
-    def forward(self, g, in_feat):
+    def forward(self, g, in_feats):
         """
         The forward function computes the output of the model.
 
@@ -71,13 +74,13 @@ def forward(self, g, in_feat):
         Returns:
             The output of the second convolutional layer
         """
-        h = self.conv1(g, in_feat)
-        h = F.relu(h)
-        h = self.conv2(g, h)
-        # h = F.relu(h)
-        # h = self.dropout(h)
-        # h = self.conv3(g, h)
-        g.ndata["h"] = h
+        for layer in self.layers:
+            if isinstance(layer, nn.ReLU):
+                in_feats = layer(in_feats)
+            else:
+                in_feats = layer(g, in_feats)
+
+        g.ndata["h"] = in_feats
         return dgl.mean_nodes(g, "h")
 
     def fit(self, train_dataloader, learning_rate=0.01, num_epochs=20):
@@ -227,14 +230,12 @@ def fit_and_evaluate(
 
             epoch_values["Loss/test"] = test_loss
             epoch_values["Accuracy/test"] = num_correct / num_total
-            epoch_values["Layer/conv1"] = self.conv1.weight.detach()
-            epoch_values["Layer/conv2"] = self.conv2.weight.detach()
-<<<<<<< HEAD
+            # to visualise distribution of tensors
+            # epoch_values["Layer/conv1"] = self.conv1.weight.detach()
+
             for i, pg in enumerate(optimizer.param_groups):
                 epoch_values[f"LearningRate/ParGrp{i}"] = pg["lr"]
-=======
             # epoch_values["Layer/conv3"] = self.conv3.weight.detach()
->>>>>>> experimenting 3 layers
             if self.num_classes == 2:
                 epoch_values["FalseNegativeRate/test"] = num_false_negative / num_total
                 epoch_values["FalsePositiveRate/test"] = num_false_positive / num_total
diff --git a/tests/test_gnn_model.py b/tests/test_gnn_model.py
index b3509c0..f73e4ee 100644
--- a/tests/test_gnn_model.py
+++ b/tests/test_gnn_model.py
@@ -6,6 +6,7 @@
 from bird_cloud_gnn.callback import EarlyStopperCallback
 from bird_cloud_gnn.callback import TensorboardCallback
 from bird_cloud_gnn.gnn_model import GCN
+from torch import nn
 
 
 def test_gnn_model(dataset_fixture):
@@ -30,7 +31,7 @@ def test_gnn_model(dataset_fixture):
         drop_last=False,
     )
 
-    model = GCN(len(dataset_fixture.features), 16, 2)
+    model = GCN(len(dataset_fixture.features), [(16, nn.ReLU()), (2, None)])
     model.fit(train_dataloader)
     model.evaluate(test_dataloader)
 
@@ -53,13 +54,13 @@ class TestBasicBehaviour:
 
     def test_field_access(self):
         """Test field access"""
-        model = GCN(in_feats=10, h_feats=16, num_classes=2)
+        model = GCN(in_feats=10, layers_data=[(16, nn.ReLU()), (2, None)])
         assert model.in_feats == 10
-        assert model.h_feats == 16
+        assert model.name == "10-16_ReLU_16-2_"
         assert model.num_classes == 2
 
     def test_inequality(self):
         """Test inequality of created GCN classes"""
-        model1 = GCN(in_feats=10, h_feats=16, num_classes=2)
-        model2 = GCN(in_feats=15, h_feats=16, num_classes=5)
+        model1 = GCN(in_feats=10, layers_data=[(16, nn.ReLU()), (2, None)])
+        model2 = GCN(in_feats=15, layers_data=[(16, nn.ReLU()), (2, None)])
         assert model1 != model2

From 3ee23ce8df2801318e73a57a96631f022980f2dc Mon Sep 17 00:00:00 2001
From: Olga Lyashevska <olga@herenstraat.nl>
Date: Tue, 11 Jul 2023 14:30:32 +0000
Subject: [PATCH 3/6] testing dynamic layers

---
 bird_cloud_gnn/gnn_model.py | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)

diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index efe76da..75cd684 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -8,8 +8,8 @@
 from dgl.nn.pytorch.conv import GraphConv
 from torch import nn
 from torch import optim
-from tqdm import tqdm
 from torch.nn.modules import Module
+from tqdm import tqdm
 
 
 os.environ["DGLBACKEND"] = "pytorch"
@@ -50,17 +50,13 @@ def __init__(self, in_feats: int, layers_data: list):
                 ), "Each tuples should contain a size (int) and a torch.nn.modules.Module."
                 self.layers.append(activation)
                 self.name = self.name + "ReLU_"
-            self.num_classes = size  # the last size should correspond to the number of classes were predicting
+            self.num_classes = (
+                size
+            )  # the last size should correspond to the number of classes were predicting
 
     def oneline_description(self):
         """Description of the model to uniquely identify it in logs"""
-        return "-".join(
-            [
-                "in_",
-                f"{self.name}",
-                "mean-out",
-            ]
-        )
+        return "-".join(["in_", f"{self.name}", "mean-out"])
 
     def forward(self, g, in_feats):
         """
@@ -263,10 +259,7 @@ def infer(self, dataset, batch_size=1024):
         """
         self.eval()
         dataloader = GraphDataLoader(
-            shuffle=False,
-            dataset=dataset,
-            batch_size=batch_size,
-            drop_last=False,
+            shuffle=False, dataset=dataset, batch_size=batch_size, drop_last=False
         )
         labels = np.array([])
         for batched_graph, _ in dataloader:

From 2564fbaca043f18c05430342dd7069fb48e4a6be Mon Sep 17 00:00:00 2001
From: Olga Lyashevska <olga@herenstraat.nl>
Date: Tue, 11 Jul 2023 16:45:40 +0200
Subject: [PATCH 4/6] Cross validation tests

---
 bird_cloud_gnn/cross_validation.py | 16 +++++++---------
 bird_cloud_gnn/gnn_model.py        | 10 +++-------
 tests/test_cross_validation.py     | 11 +++++++----
 tests/test_gnn_model.py            | 12 +++++++-----
 4 files changed, 24 insertions(+), 25 deletions(-)

diff --git a/bird_cloud_gnn/cross_validation.py b/bird_cloud_gnn/cross_validation.py
index b7bf797..91c1277 100644
--- a/bird_cloud_gnn/cross_validation.py
+++ b/bird_cloud_gnn/cross_validation.py
@@ -2,11 +2,11 @@
 """
 import numpy as np
 import pandas as pd
+from bird_cloud_gnn.gnn_model import GCN
 from dgl.dataloading import GraphDataLoader
 from sklearn.model_selection import StratifiedKFold
 from torch.utils.data.sampler import SubsetRandomSampler
 from tqdm import tqdm
-from bird_cloud_gnn.gnn_model import GCN
 
 
 def get_dataloaders(dataset, train_idx, test_idx, batch_size):
@@ -32,7 +32,7 @@ def get_dataloaders(dataset, train_idx, test_idx, batch_size):
 # pylint: disable=too-many-arguments, too-many-locals
 def kfold_evaluate(
     dataset,
-    h_feats=16,
+    layers_data,
     n_splits=5,
     learning_rate=0.01,
     num_epochs=100,
@@ -43,7 +43,7 @@ def kfold_evaluate(
 
     Args:
         dataset (RadarDataset): The dataset
-        h_feats (int, optional): The number of hidden features of the model
+        layers_data (list): The list of input size and activation
         n_splits (int, optional): Number of folds. Defaults to 5.
         learning_rate (float, optional): Learning rate. Defaults to 0.01.
         num_epochs (int, optional): Training epochs. Defaults to 20.
@@ -62,8 +62,7 @@ def kfold_evaluate(
 
         model = GCN(
             in_feats=len(dataset.features),
-            h_feats=h_feats,
-            num_classes=2,
+            layers_data=layers_data,
         )
         model.fit(train_dataloader, learning_rate=learning_rate, num_epochs=num_epochs)
 
@@ -76,7 +75,7 @@ def kfold_evaluate(
 
 def leave_one_origin_out_evaluate(
     dataset,
-    h_feats=16,
+    layers_data,
     learning_rate=0.01,
     num_epochs=100,
     batch_size=512,
@@ -88,7 +87,7 @@ def leave_one_origin_out_evaluate(
 
     Args:
         dataset (RadarDataset): The dataset.
-        h_feats (int, optional): The number of hidden features of the model
+        layers_data (list): The list of input size and activation
         n_splits (int, optional): Number of folds. Defaults to 5.
         learning_rate (float, optional): Learning rate. Defaults to 0.01.
         num_epochs (int, optional): Training epochs. Defaults to 20.
@@ -110,8 +109,7 @@ def leave_one_origin_out_evaluate(
 
         model = GCN(
             in_feats=len(dataset.features),
-            h_feats=h_feats,
-            num_classes=2,
+            layers_data=layers_data,
         )
         model.fit(train_dataloader, learning_rate=learning_rate, num_epochs=num_epochs)
 
diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index 75cd684..96cab62 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -1,17 +1,15 @@
 """Module for creating GCN class"""
 
 import os
+
 import dgl
 import numpy as np
-
 from dgl.dataloading import GraphDataLoader
 from dgl.nn.pytorch.conv import GraphConv
-from torch import nn
-from torch import optim
+from torch import nn, optim
 from torch.nn.modules import Module
 from tqdm import tqdm
 
-
 os.environ["DGLBACKEND"] = "pytorch"
 
 
@@ -50,9 +48,7 @@ def __init__(self, in_feats: int, layers_data: list):
                 ), "Each tuples should contain a size (int) and a torch.nn.modules.Module."
                 self.layers.append(activation)
                 self.name = self.name + "ReLU_"
-            self.num_classes = (
-                size
-            )  # the last size should correspond to the number of classes were predicting
+            self.num_classes = size  # the last size should correspond to the number of classes were predicting
 
     def oneline_description(self):
         """Description of the model to uniquely identify it in logs"""
diff --git a/tests/test_cross_validation.py b/tests/test_cross_validation.py
index 6062338..ea6fe9c 100644
--- a/tests/test_cross_validation.py
+++ b/tests/test_cross_validation.py
@@ -1,6 +1,9 @@
 """Tests for cross_validation"""
-from bird_cloud_gnn.cross_validation import kfold_evaluate
-from bird_cloud_gnn.cross_validation import leave_one_origin_out_evaluate
+from bird_cloud_gnn.cross_validation import (
+    kfold_evaluate,
+    leave_one_origin_out_evaluate,
+)
+from torch import nn
 
 
 def test_kfold_evaluate(dataset_fixture):
@@ -8,7 +11,7 @@ def test_kfold_evaluate(dataset_fixture):
 
     kfold_evaluate(
         dataset_fixture,
-        h_feats=32,
+        layers_data=[(32, nn.ReLU()), (2, None)],
     )
 
 
@@ -17,5 +20,5 @@ def test_leave_one_out_evaluate(dataset_fixture):
 
     leave_one_origin_out_evaluate(
         dataset_fixture,
-        h_feats=32,
+        layers_data=[(32, nn.ReLU()), (2, None)],
     )
diff --git a/tests/test_gnn_model.py b/tests/test_gnn_model.py
index f73e4ee..60c3a22 100644
--- a/tests/test_gnn_model.py
+++ b/tests/test_gnn_model.py
@@ -1,12 +1,14 @@
 """Tests for gnn_model module"""
 import torch
-from dgl.dataloading import GraphDataLoader
-from torch.utils.data.sampler import SubsetRandomSampler
-from bird_cloud_gnn.callback import CombinedCallback
-from bird_cloud_gnn.callback import EarlyStopperCallback
-from bird_cloud_gnn.callback import TensorboardCallback
+from bird_cloud_gnn.callback import (
+    CombinedCallback,
+    EarlyStopperCallback,
+    TensorboardCallback,
+)
 from bird_cloud_gnn.gnn_model import GCN
+from dgl.dataloading import GraphDataLoader
 from torch import nn
+from torch.utils.data.sampler import SubsetRandomSampler
 
 
 def test_gnn_model(dataset_fixture):

From c3c21edffd3e264048ebbbf5f845d7c20cb7afcc Mon Sep 17 00:00:00 2001
From: Bart <bart@myotis>
Date: Tue, 11 Jul 2023 18:43:21 +0200
Subject: [PATCH 5/6] optionally allow for differen activations

---
 bird_cloud_gnn/cross_validation.py |  2 +-
 bird_cloud_gnn/gnn_model.py        |  9 +++++----
 tests/test_cross_validation.py     |  6 ++----
 tests/test_gnn_model.py            | 17 +++++++++++------
 4 files changed, 19 insertions(+), 15 deletions(-)

diff --git a/bird_cloud_gnn/cross_validation.py b/bird_cloud_gnn/cross_validation.py
index 91c1277..6bdc7c9 100644
--- a/bird_cloud_gnn/cross_validation.py
+++ b/bird_cloud_gnn/cross_validation.py
@@ -2,11 +2,11 @@
 """
 import numpy as np
 import pandas as pd
-from bird_cloud_gnn.gnn_model import GCN
 from dgl.dataloading import GraphDataLoader
 from sklearn.model_selection import StratifiedKFold
 from torch.utils.data.sampler import SubsetRandomSampler
 from tqdm import tqdm
+from bird_cloud_gnn.gnn_model import GCN
 
 
 def get_dataloaders(dataset, train_idx, test_idx, batch_size):
diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index 96cab62..401d9f3 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -1,15 +1,16 @@
 """Module for creating GCN class"""
 
 import os
-
 import dgl
 import numpy as np
 from dgl.dataloading import GraphDataLoader
 from dgl.nn.pytorch.conv import GraphConv
-from torch import nn, optim
+from torch import nn
+from torch import optim
 from torch.nn.modules import Module
 from tqdm import tqdm
 
+
 os.environ["DGLBACKEND"] = "pytorch"
 
 
@@ -47,7 +48,7 @@ def __init__(self, in_feats: int, layers_data: list):
                     activation, Module
                 ), "Each tuples should contain a size (int) and a torch.nn.modules.Module."
                 self.layers.append(activation)
-                self.name = self.name + "ReLU_"
+                self.name = self.name + repr(activation).split("(", 1)[0] + "_"
             self.num_classes = size  # the last size should correspond to the number of classes were predicting
 
     def oneline_description(self):
@@ -67,7 +68,7 @@ def forward(self, g, in_feats):
             The output of the second convolutional layer
         """
         for layer in self.layers:
-            if isinstance(layer, nn.ReLU):
+            if isinstance(layer, (nn.ReLU, nn.LeakyReLU, nn.ELU)):
                 in_feats = layer(in_feats)
             else:
                 in_feats = layer(g, in_feats)
diff --git a/tests/test_cross_validation.py b/tests/test_cross_validation.py
index ea6fe9c..954e664 100644
--- a/tests/test_cross_validation.py
+++ b/tests/test_cross_validation.py
@@ -1,9 +1,7 @@
 """Tests for cross_validation"""
-from bird_cloud_gnn.cross_validation import (
-    kfold_evaluate,
-    leave_one_origin_out_evaluate,
-)
 from torch import nn
+from bird_cloud_gnn.cross_validation import kfold_evaluate
+from bird_cloud_gnn.cross_validation import leave_one_origin_out_evaluate
 
 
 def test_kfold_evaluate(dataset_fixture):
diff --git a/tests/test_gnn_model.py b/tests/test_gnn_model.py
index 60c3a22..aa95440 100644
--- a/tests/test_gnn_model.py
+++ b/tests/test_gnn_model.py
@@ -1,14 +1,12 @@
 """Tests for gnn_model module"""
 import torch
-from bird_cloud_gnn.callback import (
-    CombinedCallback,
-    EarlyStopperCallback,
-    TensorboardCallback,
-)
-from bird_cloud_gnn.gnn_model import GCN
 from dgl.dataloading import GraphDataLoader
 from torch import nn
 from torch.utils.data.sampler import SubsetRandomSampler
+from bird_cloud_gnn.callback import CombinedCallback
+from bird_cloud_gnn.callback import EarlyStopperCallback
+from bird_cloud_gnn.callback import TensorboardCallback
+from bird_cloud_gnn.gnn_model import GCN
 
 
 def test_gnn_model(dataset_fixture):
@@ -66,3 +64,10 @@ def test_inequality(self):
         model1 = GCN(in_feats=10, layers_data=[(16, nn.ReLU()), (2, None)])
         model2 = GCN(in_feats=15, layers_data=[(16, nn.ReLU()), (2, None)])
         assert model1 != model2
+
+    def test_inequality_activation(self):
+        """Test inequality of created GCN classes with different activation"""
+        model1 = GCN(in_feats=10, layers_data=[(16, nn.ReLU()), (2, None)])
+        model2 = GCN(in_feats=10, layers_data=[(16, nn.ELU()), (2, None)])
+        assert model1 != model2
+        assert model2.name == "10-16_ELU_16-2_"

From ed18edbe232d3fdef95602f150752683fe24c2c6 Mon Sep 17 00:00:00 2001
From: Bart <bart@myotis>
Date: Tue, 11 Jul 2023 19:08:08 +0200
Subject: [PATCH 6/6] restore layer info in tensor board

---
 bird_cloud_gnn/gnn_model.py | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/bird_cloud_gnn/gnn_model.py b/bird_cloud_gnn/gnn_model.py
index 401d9f3..4698139 100644
--- a/bird_cloud_gnn/gnn_model.py
+++ b/bird_cloud_gnn/gnn_model.py
@@ -223,12 +223,13 @@ def fit_and_evaluate(
 
             epoch_values["Loss/test"] = test_loss
             epoch_values["Accuracy/test"] = num_correct / num_total
-            # to visualise distribution of tensors
-            # epoch_values["Layer/conv1"] = self.conv1.weight.detach()
 
             for i, pg in enumerate(optimizer.param_groups):
                 epoch_values[f"LearningRate/ParGrp{i}"] = pg["lr"]
-            # epoch_values["Layer/conv3"] = self.conv3.weight.detach()
+            # to visualise distribution of tensors
+            for i, layer in enumerate(self.layers):
+                if not isinstance(layer, (nn.ReLU, nn.LeakyReLU, nn.ELU)):
+                    epoch_values[f"Layer/conv{i}"] = layer.weight.detach()
             if self.num_classes == 2:
                 epoch_values["FalseNegativeRate/test"] = num_false_negative / num_total
                 epoch_values["FalsePositiveRate/test"] = num_false_positive / num_total