docs

examples/train-web/readme.md

@@ -6,7 +6,9 @@ For example, run `cargo install --version 0.2.88 wasm-bindgen-cli --force`. The
 
 Install [PNPM](https://pnpm.io/).
 
-The [`postinstall.sh`](./web/postinstall.sh) script expects the mnist database to be at `~/.cache/burn-dataset/mnist.db`. Running `guide` will generate this file. Alternatively, you can download it from [Hugging Face](https://huggingface.co/datasets/mnist).
+Install [cargo-watch](https://crates.io/crates/cargo-watch).
+
+The [`postinstall.sh`](./web/postinstall.sh) script expects the mnist database to be at `~/.cache/burn-dataset/mnist.db`. Running `burn/examples/guide` will generate this file. Alternatively, you can download it from [Hugging Face](https://huggingface.co/datasets/mnist).
 
 Then in separate terminals:
 

examples/train-web/train/src/mnist.rs

@@ -71,6 +71,8 @@ impl Dataset<MNISTItem> for MNISTDataset {
 impl MNISTDataset {
     /// Creates a new dataset.
     pub fn new(labels: &[u8], images: &[u8], lengths: &[u16]) -> Self {
+        // Decoding is here.
+        // Encoding is done at `examples/train-web/web/src/train.ts`.
         debug_assert!(labels.len() == lengths.len());
         let mut start = 0 as usize;
         let raws = labels

examples/train-web/web/src/train.ts

@@ -34,6 +34,12 @@ export function setupTrain(element: HTMLInputElement) {
 }
 
 async function loadSqliteAndRun(ab: ArrayBuffer) {
+	// Images are an array of arrays.
+	// We can't send an array of arrays to Wasm.
+	// So instead we merge images into a single large array and
+	// use another array, `lengths`, to keep track of the image size.
+	// Encoding is done here.
+	// Decoding is done at `burn/examples/train-web/train/src/mnist.rs`.
 	const trainImages: Uint8Array[] = []
 	const trainLabels: number[] = []
 	const trainLengths: number[] = []