Just one of the things I'm learning. https://github.com/hchiam/learning
Using Universal-Sentence-Encoder (USE) and UMAP to attempt to graph sentences with semantically-similar meanings close together. UMAP is conceptually similar to PCA in that it also reduces dimensions, but UMAP is stochastic to help with speed of calculation for ML purposes, so the output of UMAP isn't always the same.
Get UMAP output coordinates with umap-js and visualization powered by chart.js.
Interactive explanation of UMAP: https://pair-code.github.io/understanding-umap
Great explanation of SNE vs t-SNE (vs UMAP - which sounds better than t-SNE): https://towardsdatascience.com/visualizing-your-embeddings-4c79332581a9 - basically UMAP is better than t-SNE is better than SNE (which is related to the order they were invented). My other repo that uses tfjs-tsne can be found here: https://github.com/hchiam/learning-tfjs-tsne
Learn more about TensorFlow at https://github.com/hchiam/learning-tensorflow
Repo created from this template for further experimentation: https://github.com/hchiam/comment-analysis
Run yarn global add parcel
or npm install -g parcel
, and then:
Using yarn
:
git clone https://github.com/hchiam/learning-tfjs-umap.git && cd learning-tfjs-umap && yarn;
yarn dev;
Or with npm
:
git clone https://github.com/hchiam/learning-tfjs-umap.git && cd learning-tfjs-umap && npm install;
npm run dev;
First run yarn dev
to create the /dist folder, and then run this:
yarn deploy