Guided manifold learning and semi-supervised visualization with natural out-of-sample extension based on random forest proximities and diffusion geometry-regularized autoencoder architecture.