Upload 6 files

README.md

@@ -0,0 +1,3 @@
+# Assignment 4 - AI417
+
+This repository contains my trained models (DummyNet and VanillaNet) for Assignment 4.

dummy-weights.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a832d7797004f5ea77e8f3d791cfbf9509253971ae731b3fbfb9d985a88a50b1
+size 824029

dummy.py

@@ -0,0 +1,37 @@
+import random
+import numpy as np
+import torch as tr
+
+seed = 4320232                    # <<<<<<<<<<<<<<<< Your UPM ID Goes Here
+random.seed(seed)
+np.random.seed(seed)
+tr.manual_seed(seed)
+
+class SimpleFeedForwardNet(tr.nn.Module):
+
+  def __init__(self):
+    super().__init__() 
+    self.linear1 = tr.nn.Linear(784, 256, bias=True)       # 784 -> 16 - Layer 1 -- Affine Transformation (Linear with Bias)
+    self.linear2 = tr.nn.Linear(256, 16, bias=True)       # 16 -> 16  - Layer 2 -- Linear Transformation (no bias)
+    self.linear3 = tr.nn.Linear(16, 10, bias=True)        # 16 -> 10  - Layer 3 -- Affine Transformation (Linear with Bias)
+    #self.init_weights()
+
+  def init_weights(self):
+    tr.nn.init.zeros_(self.linear1.weight)
+    tr.nn.init.zeros_(self.linear2.weight)
+    tr.nn.init.zeros_(self.linear3.weight)
+    tr.nn.init.zeros_(self.linear1.bias)
+    tr.nn.init.ones_(self.linear2.bias)
+    tr.nn.init.ones_(self.linear3.bias)
+  
+  def forward(self, x):
+    x = self.linear1(x)
+    x = self.linear2(x)
+    x = self.linear3(x)
+    return x
+
+model = SimpleFeedForwardNet()                                          # Architecture
+optimizer = tr.optim.SGD(model.parameters(), lr=0.02, maximize=False)   # Optimizer
+
+
+loss_fn = tr.nn.CrossEntropyLoss()                                      # Objective Function [DO NOT CHANGE !]

hubconf.py

@@ -0,0 +1,28 @@
+dependencies = ['torch']
+
+import torch
+import os
+
+def DummyNet(pretrained=True, **kwargs):
+    """Entry point for the DummyNet model."""
+    from dummy import SimpleFeedForwardNet
+    model = SimpleFeedForwardNet(**kwargs)
+    
+    if pretrained:
+        hub_dir = os.path.dirname(os.path.abspath(__file__))
+        weight_path = os.path.join(hub_dir, 'dummy-weights.bin')
+        model.load_state_dict(torch.load(weight_path))
+        
+    return model
+
+def VanillaNet(pretrained=True, **kwargs):
+    """Entry point for the VanillaNet model."""
+    from vanilla import SimpleFeedForwardNet
+    model = SimpleFeedForwardNet(**kwargs)
+    
+    if pretrained:
+        hub_dir = os.path.dirname(os.path.abspath(__file__))
+        weight_path = os.path.join(hub_dir, 'vanilla-weight.bin')
+        model.load_state_dict(torch.load(weight_path))
+        
+    return model

vanilla-weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee546974cd4bc192ede04b3c0c36d7c783de5eb68c89813243a76eb055aae9fc
+size 816675

vanilla.py

@@ -0,0 +1,36 @@
+import random
+import numpy as np
+import torch as tr
+
+seed = 4320232                    # <<<<<<<<<<<<<<<< Your UPM ID Goes Here
+random.seed(seed)
+np.random.seed(seed)
+tr.manual_seed(seed)
+
+class SimpleFeedForwardNet(tr.nn.Module):
+
+  def __init__(self):
+    super().__init__() 
+    self.linear1 = tr.nn.Linear(784, 256)
+    self.activation1 = tr.nn.ReLU()  # 784 -> 16 - Layer 1 -- Affine Transformation (Linear with Bias)
+    self.linear2 = tr.nn.Linear(256, 10)       # 16 -> 16  - Layer 2 -- Linear Transformation (no bias)
+         # 16 -> 10  - Layer 3 -- Affine Transformation (Linear with Bias)
+    #self.init_weights()
+
+  def init_weights(self):
+    tr.nn.init.kaiming_uniform_(self.linear1.weight, nonlinearity='relu')
+    tr.nn.init.constant_(self.linear1.bias, 0)
+    tr.nn.init.kaiming_uniform_(self.linear2.weight, nonlinearity='relu')
+    tr.nn.init.constant_(self.linear2.bias, 0)    
+  
+  def forward(self, x):
+    x = self.linear1(x)
+    x = self.activation1(x)
+    x = self.linear2(x)
+    return x
+
+model = SimpleFeedForwardNet()                                          # Architecture
+optimizer = tr.optim.Adam(model.parameters(), lr=0.001)   # Optimizer
+
+
+loss_fn = tr.nn.CrossEntropyLoss()                                      # Objective Function [DO NOT CHANGE !]