Fix CNN Algo

Author: ai-nerdy

machine_learning/cnn.py

@@ -12,11 +12,10 @@
 """
 
 import numpy as np
-from typing import Tuple
 
 
 class SimpleCNN:
-    def __init__(self, input_shape: Tuple[int, int, int], num_classes: int) -> None:
+    def __init__(self, input_shape: tuple[int, int, int], num_classes: int) -> None:
         """
         Initialize a simple CNN model.
 
@@ -26,42 +25,43 @@ def __init__(self, input_shape: Tuple[int, int, int], num_classes: int) -> None:
         """
         self.input_shape = input_shape
         self.num_classes = num_classes
-        self.filters = np.random.randn(8, input_shape[0], 3, 3) * 0.1  # 8 filters
-        self.fc_weights = np.random.randn(8 * 26 * 26, num_classes) * 0.1
+        rng = np.random.default_rng()
+        self.filters = rng.normal(0, 0.1, size=(8, input_shape[0], 3, 3))  # 8 filters
+        self.fc_weights = rng.normal(0, 0.1, size=(8 * 26 * 26, num_classes))
 
-    def relu(self, x: np.ndarray) -> np.ndarray:
-        """Apply ReLU activation."""
-        return np.maximum(0, x)
+    def relu(self, feature_map: np.ndarray) -> np.ndarray:
+        """Apply ReLU activation to the feature map."""
+        return np.maximum(0, feature_map)
 
-    def convolve(self, x: np.ndarray, filters: np.ndarray) -> np.ndarray:
-        """Apply convolution operation."""
-        batch, height, width = x.shape
+    def convolve(self, input_tensor: np.ndarray, filters: np.ndarray) -> np.ndarray:
+        """Apply convolution operation to the input tensor."""
+        _, height, width = input_tensor.shape
         num_filters, _, fh, fw = filters.shape
         output = np.zeros((num_filters, height - fh + 1, width - fw + 1))
 
         for f in range(num_filters):
             for i in range(height - fh + 1):
                 for j in range(width - fw + 1):
-                    region = x[:, i:i + fh, j:j + fw]
+                    region = input_tensor[:, i:i + fh, j:j + fw]
                     output[f, i, j] = np.sum(region * filters[f])
         return output
 
-    def flatten(self, x: np.ndarray) -> np.ndarray:
-        """Flatten the feature map."""
-        return x.reshape(-1)
+    def flatten(self, feature_map: np.ndarray) -> np.ndarray:
+        """Flatten the feature map into a 1D array."""
+        return feature_map.reshape(-1)
 
-    def forward(self, x: np.ndarray) -> np.ndarray:
+    def forward(self, input_tensor: np.ndarray) -> np.ndarray:
         """
         Forward pass through the CNN.
 
         Args:
-            x: Input image of shape (channels, height, width)
+            input_tensor: Input image of shape (channels, height, width)
 
         Returns:
             Output logits of shape (num_classes,)
         """
-        conv_out = self.convolve(x, self.filters)
+        conv_out = self.convolve(input_tensor, self.filters)
         activated = self.relu(conv_out)
         flattened = self.flatten(activated)
         logits = flattened @ self.fc_weights
-        return logits
+        return logits