Learning And Inference For Adaptable Manipulation Planning