If you plan to work with Nvidia's CUDA framework right away to do some GPU-based computations (neural networks primarily), you should stick with GeForce 750M laptop.
However, 750M's memory is only 2 GB (as the R370X is), so you will not be able to fit any big model into your GPU either way and will be forced to use either cloud instances or buy a separate GPU and somehow connect it to your Mac so the choice is really not too much relevant.
If you don't have any immediate plans to dive into CUDA-based computations, buy either the cheapest option (750M Macs are previous generation, so you will buy them refurbished or used and that may save some money), or the newest (which is R370X) since it will be brand new and you will enjoy warranty period at its fullest.
Also, Apple did release the Metal framework for native shader programming in OS X, which allows to use literally any supported GPU from OS X to program your computations, and there are some frameworks that use that possibility (Memkite for example) - that will give you deep learning and computer vision on any mentioned GPU (and even on built-in Iris Pro which is built into the CPU). But the support of Metal is in its early stage, so there are not too many frameworks and projects using it for machine learning (though, you could pioneer in that).