einops VS horovod

Thanks for the pointer. I can believe that a language that looks so different will find that different patterns and primitives are natural for it.

My experience from writing a lot of array-based code in NumPy/Matlab is that broadcasting absolutely has made it easier to write my code in those ecosystems. Axes of length 1 have often been in the right places already, or have been easy to insert. It's of course possible to create a big mess in any language; it seems likely that the NumPy code you saw could have been neater too.

In machine learning there can be many array dimensions floating around: batch-dims, sequence and/or channel-dims, weight matrices, and so on. It can be necessary to expand two or more dimensions, and/or line up dimensions quite carefully. Einops[1] has emerged from that community as a tool to succinctly express many operations that involve lots of array dimensions. You're likely to bump into more and more people who've used it, and again it seems there's some overlap with what Rank does. (And again, you'll see uses of Einops in the wild that are unnecessarily convoluted.)

[1] https://einops.rocks/ -- It works with all of the existing major array-based frameworks for Python (NumPy/PyTorch/Jax/etc), and the emerging array API standard for Python.

this is definitely what you're looking for: https://github.com/arogozhnikov/einops

Einops all the things! https://einops.rocks/

It depends on what are their main purposes. I know some figures who have done an amazing job in this field but never because of publications, e.g. https://github.com/lucidrains and https://github.com/rwightman, https://einops.rocks/

On tsalib's warp: this is very similar to einops. I think it might even be slightly more general. However, I'm honestly not sure to what extent tsalib is still maintained, as it looks like the most recent commit was about two years ago. (Which is a shame.)

Also see Einops: https://github.com/arogozhnikov/einops, which uses a einsum-like notation for various tensor operations used in deep learning.

https://einops.rocks/pytorch-examples.html shows how it can be used to implement various neural network architectures in a more simplified manor.

This combined with something like einops [1] ( intuitive reshaping library) can be a huge time saver.

[1] https://github.com/arogozhnikov/einops

einops: just look at the pretty visual GIF and be amazed

Broke: using Horovod

And in case scaling is really important to you. May I suggest you look into Horovod?

Training is not the same as inference (doing the segmentation), so that scale is probably off by a lot. One or two orders of magnitude just depending on the specifics of what hardware you're running on, and your training and eval dataset would be several orders of magnitude smaller. FAANGs would parallelize that training as well (don't remember if UNet is inherently parallelizable for training) via their internal equivalent of Horovod, so they'll do a GPU-month worth of training in less than a day.

[[email protected]] match_arg (utils/args/args.c:163): unrecognized argument quiet [[email protected]] HYDU_parse_array (utils/args/args.c:178): argument matching returned error [[email protected]] parse_args (ui/mpich/utils.c:1639): error parsing input array [[email protected]] HYD_uii_mpx_get_parameters (ui/mpich/utils.c:1691): unable to parse user arguments [[email protected]] main (ui/mpich/mpiexec.c:127): error parsing parameters I believe this is due to mpich being installed: https://github.com/horovod/horovod/issues/1637

Check out Horovod - https://github.com/horovod/horovod

If you just want to run tensorflow or pytorch for a Jupyter notebook, setting the environment shouldn't be difficult. I know that AWS has a marketplace of preconfigured images. However, you can go as advanced as setting up a cluster of gpu-equipped nodes to setup Horovod (https://github.com/horovod/horovod) to do distributed machine learning. Yes, there's a learning curve, but you cannot acquire this skillet any other way.

I'm the maintainer for an open source project called Horovod that allows you to distribute deep learning training (e.g., TensorFlow) on platforms like Spark.

You'll want to use horovod to run keras in a distributed system. Then use Slurm to manage the cluster and run the job.