Published
Edited
May 18, 2022
7 stars
height = width
{
const features = ds.mnist.map((instance) => instance.features);
const { grids, imageHeight, imageWidth } = generateGrid({
width: width,
height: height,
rows: 20,
cols: 20
});
const tensorFeatures = tf.tensor(features);
let iteration = 0;
for await (const assignments of tsneGrid(tensorFeatures, grids)) {
const useableFormat = assignments.map((ass, i) => {
const instance = ds.mnist[ass];
return { id: instance.filename, grid: grids[i] };
});
svg
.selectAll("image")
.data(useableFormat)
.join("image")
.attr("x", (d) => d.grid[0])
.attr("y", (d) => d.grid[1])
.attr("width", imageWidth)
.attr("height", imageHeight)
.attr("href", (d) => `${settings.href}/${d.id}`);
progressSvg
.select("rect")
.attr("width", (iteration++ / settings.iterations.tsne) * width);
}
}
async function* tsneGrid(data, gridPoints) {
const tsne = tf.tsne.tsne(data);
await tsne.iterateKnn(settings.iterations.knn);
for (let i = 0; i < settings.iterations.tsne; i++) {
await tsne.iterate();
const featuresTensor = await tsne.coordinates();
const featuresFlattened = await featuresTensor.data();
await featuresTensor.dispose();
const features2D = reshape2D(featuresFlattened, [data.shape[0], 2]);
const transformFeatures2D = transformToGridSpace(features2D);
const assignments = linearOptimizer.assign({
points: transformFeatures2D,
assignTo: gridPoints
});
yield assignments;
}
}
d3 = require("d3@6")
One platform to build and deploy the best data apps
Experiment and prototype by building visualizations in live JavaScript notebooks. Collaborate with your team and decide which concepts to build out.
Use Observable Framework to build data apps locally. Use data loaders to build in any language or library, including Python, SQL, and R.
Seamlessly deploy to Observable. Test before you ship, use automatic deploy-on-commit, and ensure your projects are always up-to-date.