Skip to content
Deepbox
27intermediate

CNN Layers

Runs the convolution and pooling layers Conv1d, Conv2d, MaxPool2d and AvgPool2d on small inputs and prints the output shapes. It also shows padding: "same" and padding: "valid", and chains layers with Sequential.

What you will learn

  • Use deepbox/ndarray for tensor, noGrad.
  • Use deepbox/nn for Conv1d, Conv2d, MaxPool2d, AvgPool2d, Sequential.
  • Runs the convolution and pooling layers Conv1d, Conv2d, MaxPool2d and AvgPool2d on small inputs and prints the output shapes. It also shows padding: "same" and padding: "valid", and chains layers with Sequential.

Source

index.ts
/** * Example 27: Convolutional Neural Network Layers * * Conv1d, Conv2d, MaxPool2d and AvgPool2d on small inputs, with the output * shapes printed. Convolutions are the building block of image and signal models. * * The forward passes below are inference only, so they run inside noGrad(). * Without it, a layer with trainable weights returns a GradTensor that tracks * the weights. Both kinds of tensor have .shape and .toString(). */ import { noGrad, tensor } from "deepbox/ndarray";import { AvgPool2d, Conv1d, Conv2d, MaxPool2d, ReLU, Sequential } from "deepbox/nn"; console.log("=== Convolutional Neural Network Layers ===\n"); // ---------------------------------------------------------------------------// Part 1: Conv1d (1D convolution for sequence and signal data)// ---------------------------------------------------------------------------console.log("--- Part 1: Conv1d ---"); // Conv1d expects input of shape (batch, inChannels, length)const conv1d = new Conv1d(1, 4, 3, { padding: 1 });console.log("Conv1d(in=1, out=4, kernel=3, padding=1)"); const signal = tensor([[[1, 2, 3, 4, 5, 6, 7, 8]]]);console.log(`Input shape:  [${signal.shape.join(", ")}]`); const conv1dOut = noGrad(() => conv1d.forward(signal));console.log(`Output shape: [${conv1dOut.shape.join(", ")}]`);console.log("  4 output channels from 1 input channel"); // padding: "same" keeps the length for any kernel size, "valid" adds no paddingconst sameConv = new Conv1d(1, 4, 5, { padding: "same" });const validConv = new Conv1d(1, 4, 5, { padding: "valid" });console.log(  `kernel=5, padding="same":  [${noGrad(() => sameConv.forward(signal)).shape.join(", ")}]`);console.log(  `kernel=5, padding="valid": [${noGrad(() => validConv.forward(signal)).shape.join(", ")}]\n`); // ---------------------------------------------------------------------------// Part 2: Conv2d (2D convolution for image data)// ---------------------------------------------------------------------------console.log("--- Part 2: Conv2d ---"); // Conv2d expects input of shape (batch, inChannels, height, width)const conv2d = new Conv2d(1, 4, 2, { bias: false });console.log("Conv2d(in=1, out=4, kernel=2x2, bias=false)"); const image = tensor([  [    [      [1, 2, 3, 4],      [5, 6, 7, 8],      [9, 10, 11, 12],      [13, 14, 15, 16],    ],  ],]);console.log(`Input shape:  [${image.shape.join(", ")}]`);const conv2dOut = noGrad(() => conv2d.forward(image));console.log(`Output shape: [${conv2dOut.shape.join(", ")}]  (4 - 2 + 1 = 3 along each side)`); const conv2dParams = Array.from(conv2d.parameters()).length;console.log(`Parameters: ${conv2dParams} (weight only, no bias)\n`); // ---------------------------------------------------------------------------// Part 3: MaxPool2d (downsampling with max pooling)// ---------------------------------------------------------------------------console.log("--- Part 3: MaxPool2d ---"); const poolInput = tensor([  [    [      [1, 2],      [3, 4],    ],  ],]);const maxPool = new MaxPool2d(2, { stride: 2 });console.log("MaxPool2d(kernel=2, stride=2)");console.log(`Input shape:  [${poolInput.shape.join(", ")}]`); // Pooling layers have no parameters, so the result is a plain tensorconst pooled = maxPool.forward(poolInput);console.log(`Output shape: [${pooled.shape.join(", ")}]`);console.log(`Output value: ${pooled.toString()}`);console.log("  Each 2x2 window is replaced by its maximum\n"); // ---------------------------------------------------------------------------// Part 4: AvgPool2d (downsampling with average pooling)// ---------------------------------------------------------------------------console.log("--- Part 4: AvgPool2d ---"); const avgPool = new AvgPool2d(2, { stride: 2 });console.log("AvgPool2d(kernel=2, stride=2)"); const avgPooled = avgPool.forward(poolInput);console.log(`Output shape: [${avgPooled.shape.join(", ")}]`);console.log(`Output value: ${avgPooled.toString()}`);console.log("  Each 2x2 window is replaced by its mean\n"); // ---------------------------------------------------------------------------// Part 5: A small Conv1d pipeline with Sequential// ---------------------------------------------------------------------------console.log("--- Part 5: Sequential Conv1d Pipeline ---"); const cnn = new Sequential(  new Conv1d(1, 4, 3, { padding: 1 }),  new ReLU(),  new Conv1d(4, 8, 3, { padding: 1 }),  new ReLU()); console.log("Sequential 1D CNN:");console.log("  Conv1d(1->4, k=3) -> ReLU -> Conv1d(4->8, k=3) -> ReLU"); const cnnInput = tensor([[[1, 2, 3, 4, 5, 6]]]);const cnnOutput = noGrad(() => cnn.forward(cnnInput));console.log(`Input shape:  [${cnnInput.shape.join(", ")}]`);console.log(`Output shape: [${cnnOutput.shape.join(", ")}]`); const paramCount = Array.from(cnn.parameters()).length;console.log(`Parameter tensors: ${paramCount} (weight and bias for each Conv1d)`); console.log("\n=== CNN Layers Complete ===");

Output

Console output
Console output only: input and output shapes for each layer, the pooled values, and the number of parameter tensors in a small `Sequential` CNN.