TinyTorch/tinytorch/core/cnn.py

# AUTOGENERATED! DO NOT EDIT! File to edit: ../../modules/source/06_spatial/spatial_dev.ipynb.

# %% auto 0
__all__ = ['conv2d_naive', 'Conv2D', 'flatten']

# %% ../../modules/source/06_spatial/spatial_dev.ipynb 1
import numpy as np
import os
import sys
from typing import List, Tuple, Optional
import matplotlib.pyplot as plt

# Import from the main package - try package first, then local modules
try:
    from tinytorch.core.tensor import Tensor
    from tinytorch.core.layers import Dense
    from tinytorch.core.activations import ReLU
except ImportError:
    # For development, import from local modules
    sys.path.append(os.path.join(os.path.dirname(__file__), '..', '01_tensor'))
    sys.path.append(os.path.join(os.path.dirname(__file__), '..', '02_activations'))
    sys.path.append(os.path.join(os.path.dirname(__file__), '..', '03_layers'))
    from tensor_dev import Tensor
    from activations_dev import ReLU
    from layers_dev import Dense

# %% ../../modules/source/06_spatial/spatial_dev.ipynb 2
def _should_show_plots():
    """Check if we should show plots (disable during testing)"""
    # Check multiple conditions that indicate we're in test mode
    is_pytest = (
        'pytest' in sys.modules or
        'test' in sys.argv or
        os.environ.get('PYTEST_CURRENT_TEST') is not None or
        any('test' in arg for arg in sys.argv) or
        any('pytest' in arg for arg in sys.argv)
    )
    
    # Show plots in development mode (when not in test mode)
    return not is_pytest

# %% ../../modules/source/06_spatial/spatial_dev.ipynb 6
def conv2d_naive(input: np.ndarray, kernel: np.ndarray) -> np.ndarray:
    """
    Naive 2D convolution (single channel, no stride, no padding).
    
    Args:
        input: 2D input array (H, W)
        kernel: 2D filter (kH, kW)
    Returns:
        2D output array (H-kH+1, W-kW+1)
        
    TODO: Implement the sliding window convolution using for-loops.
    
    APPROACH:
    1. Get input dimensions: H, W = input.shape
    2. Get kernel dimensions: kH, kW = kernel.shape
    3. Calculate output dimensions: out_H = H - kH + 1, out_W = W - kW + 1
    4. Create output array: np.zeros((out_H, out_W))
    5. Use nested loops to slide the kernel:
       - i loop: output rows (0 to out_H-1)
       - j loop: output columns (0 to out_W-1)
       - di loop: kernel rows (0 to kH-1)
       - dj loop: kernel columns (0 to kW-1)
    6. For each (i,j), compute: output[i,j] += input[i+di, j+dj] * kernel[di, dj]
    
    EXAMPLE:
    Input: [[1, 2, 3],     Kernel: [[1, 0],
            [4, 5, 6],              [0, -1]]
            [7, 8, 9]]
    
    Output[0,0] = 1*1 + 2*0 + 4*0 + 5*(-1) = 1 - 5 = -4
    Output[0,1] = 2*1 + 3*0 + 5*0 + 6*(-1) = 2 - 6 = -4
    Output[1,0] = 4*1 + 5*0 + 7*0 + 8*(-1) = 4 - 8 = -4
    Output[1,1] = 5*1 + 6*0 + 8*0 + 9*(-1) = 5 - 9 = -4
    
    HINTS:
    - Start with output = np.zeros((out_H, out_W))
    - Use four nested loops: for i in range(out_H): for j in range(out_W): for di in range(kH): for dj in range(kW):
    - Accumulate the sum: output[i,j] += input[i+di, j+dj] * kernel[di, dj]
    """
    ### BEGIN SOLUTION
    # Get input and kernel dimensions
    H, W = input.shape
    kH, kW = kernel.shape
    
    # Calculate output dimensions
    out_H, out_W = H - kH + 1, W - kW + 1
    
    # Initialize output array
    output = np.zeros((out_H, out_W), dtype=input.dtype)
    
    # Sliding window convolution with four nested loops
    for i in range(out_H):
        for j in range(out_W):
            for di in range(kH):
                for dj in range(kW):
                    output[i, j] += input[i + di, j + dj] * kernel[di, dj]
    
    return output
    ### END SOLUTION

# %% ../../modules/source/06_spatial/spatial_dev.ipynb 10
class Conv2D:
    """
    2D Convolutional Layer (single channel, single filter, no stride/pad).
    
    A learnable convolutional layer that applies a kernel to detect spatial patterns.
    Perfect for building the foundation of convolutional neural networks.
    """
    
    def __init__(self, kernel_size: Tuple[int, int]):
        """
        Initialize Conv2D layer with random kernel.
        
        Args:
            kernel_size: (kH, kW) - size of the convolution kernel
            
        TODO: Initialize a random kernel with small values.
        
        APPROACH:
        1. Store kernel_size as instance variable
        2. Initialize random kernel with small values
        3. Use proper initialization for stable training
        
        EXAMPLE:
        Conv2D((2, 2)) creates:
        - kernel: shape (2, 2) with small random values
        
        HINTS:
        - Store kernel_size as self.kernel_size
        - Initialize kernel: np.random.randn(kH, kW) * 0.1 (small values)
        - Convert to float32 for consistency
        """
        ### BEGIN SOLUTION
        # Store kernel size
        self.kernel_size = kernel_size
        kH, kW = kernel_size
        
        # Initialize random kernel with small values
        self.kernel = np.random.randn(kH, kW).astype(np.float32) * 0.1
        ### END SOLUTION
    
    def forward(self, x):
        """
        Forward pass: apply convolution to input tensor.
        
        Args:
            x: Input tensor (2D for simplicity)
            
        Returns:
            Output tensor after convolution
            
        TODO: Implement forward pass using conv2d_naive function.
        
        APPROACH:
        1. Extract numpy array from input tensor
        2. Apply conv2d_naive with stored kernel
        3. Return result wrapped in Tensor
        
        EXAMPLE:
        x = Tensor([[1, 2, 3], [4, 5, 6], [7, 8, 9]])  # shape (3, 3)
        layer = Conv2D((2, 2))
        y = layer(x)  # shape (2, 2)
        
        HINTS:
        - Use x.data to get numpy array
        - Use conv2d_naive(x.data, self.kernel)
        - Return Tensor(result) to wrap the result
        """
        ### BEGIN SOLUTION
        # Apply convolution using naive implementation
        result = conv2d_naive(x.data, self.kernel)
        return type(x)(result)
        ### END SOLUTION
    
    def __call__(self, x):
        """Make layer callable: layer(x) same as layer.forward(x)"""
        return self.forward(x)

# %% ../../modules/source/06_spatial/spatial_dev.ipynb 14
def flatten(x):
    """
    Flatten a 2D tensor to 1D (for connecting to Dense layers).
    
    Args:
        x: Input tensor to flatten
        
    Returns:
        Flattened tensor with batch dimension preserved
        
    TODO: Implement flattening operation.
    
    APPROACH:
    1. Get the numpy array from the tensor
    2. Use .flatten() to convert to 1D
    3. Add batch dimension with [None, :]
    4. Return Tensor wrapped around the result
    
    EXAMPLE:
    Input: Tensor([[1, 2], [3, 4]])  # shape (2, 2)
    Output: Tensor([[1, 2, 3, 4]])  # shape (1, 4)
    
    HINTS:
    - Use x.data.flatten() to get 1D array
    - Add batch dimension: result[None, :]
    - Return Tensor(result)
    """
    ### BEGIN SOLUTION
    # Flatten the tensor and add batch dimension
    flattened = x.data.flatten()
    result = flattened[None, :]  # Add batch dimension
    return type(x)(result)
    ### END SOLUTION