cnn with dropout

config.yml

@@ -40,7 +40,7 @@ Training:
   
 #Model selection
 Model:
-  Name: BidirectionalLSTM 
+  Name: CNN1D
   #choose in {LinearRegression, BidirectionalLSTM, RNN}
 
 #Model parameters selection
@@ -64,6 +64,9 @@ RNN:
   Dropout: 0.2
   Initialization: None
 
+CNN1D:
+  Initialization: None
+
 #Name of directory containing logs
 LogDir: ./logs/
 

model.py

@@ -5,8 +5,6 @@ import torch.nn.functional as F
 from torch.autograd import Function
 from torch.autograd import Variable
 
-import torch.nn as nn
-
 # Linear Regression (Feed forward Network)
 class LinearRegression(nn.Module):
     def __init__(self, cfg, input_size):
@@ -115,6 +113,50 @@ class BidirectionalLSTM(nn.Module):
         result = self.fc(out)
         return result
 
+
+# CNN
+class CNN1D(torch.nn.Module):
+
+    def __init__(self, cfg, num_inputs):
+        super(CNN1D, self).__init__()
+        self.model = torch.nn.Sequential(
+            *conv_block(num_inputs, 5, 0.01),
+            *conv_block(32, 6, 0.01)
+        )
+        self.avg_pool = torch.nn.AdaptiveAvgPool1d(1)
+
+        self.fc = nn.Sequential(
+            nn.Linear(64, 32),
+            nn.ReLU(),
+            nn.Linear(32,cfg["Dataset"]["num_days"])
+        )
+
+    def forward(self, x):
+        x = torch.transpose(x, 1, 2)
+
+        out = self.model(x)
+
+        out = self.avg_pool(out)
+
+        out = out.view([out.shape[0], -1])
+
+        out = self.fc(out)
+        out = out.view([out.shape[0], out.shape[1], 1])
+        return out
+
+def conv_block(in_channels, power, dropout_p):
+    return [
+        torch.nn.Conv1d(in_channels, 2**power, 16),
+        torch.nn.BatchNorm1d(2**power),
+        torch.nn.ReLU(),
+        torch.nn.Dropout(p=dropout_p),
+        #torch.nn.Conv1d(2**power, 2**power, 8),
+        #torch.nn.BatchNorm1d(2**power),
+        #torch.nn.ReLU(),
+        #torch.nn.Dropout(p=dropout_p),
+        torch.nn.MaxPool1d(2, stride = 1)
+    ]
+
 # Initialization
 def init_he(module):
     if type(module)==nn.Linear:

test.py

@@ -35,10 +35,10 @@ def test(model, loader, f_loss, device):
 
             # Compute the forward pass, i.e. the scores for each input image
             outputs = model(inputs)
-            print("Validation inputs :")
-            print(inputs)
-            print("Validation outputs :")
-            print(outputs)
+            #print("Validation inputs :")
+            #print(inputs)
+            #print("Validation outputs :")
+            #print(outputs)
 
             # We accumulate the exact number of processed samples
             N += inputs.shape[0]

train.py

@@ -30,6 +30,8 @@ def train(args, model, loader, f_loss, optimizer, device, log_interval = 100):
         inputs, targets = inputs.to(device), targets.to(device)
 
         # Compute the forward pass through the network up to the loss
+
+        # target's shape is (B, Num_days, 1)
         outputs = model(inputs)
         loss = f_loss(outputs, targets)
 

wandb/debug-internal.log

-run-20230204_010308-1aksu4p8/logs/debug-internal.log
\ No newline at end of file
+run-20230204_013952-5w9xw0aw/logs/debug-internal.log
\ No newline at end of file

wandb/debug.log

-run-20230204_010308-1aksu4p8/logs/debug.log
\ No newline at end of file
+run-20230204_013952-5w9xw0aw/logs/debug.log
\ No newline at end of file

wandb/latest-run

-run-20230204_010308-1aksu4p8
\ No newline at end of file
+run-20230204_013952-5w9xw0aw
\ No newline at end of file