From $t$-SNE to UMAP with contrastive learning