We found precise responses to orientations in mouse visual cortex in 50,000 neuron recordings. To understand neural recordings from more complex experiments, we developed an embedding algorithm called Rastermap, which qualitatively and quantitatively outperforms t-SNE and UMAP.