Michelangelo PyML: Uber’s Platform for Rapid Python ML Model | Hacker News Reader

Michelangelo PyML: Uber’s Platform for Rapid Python ML Model

eng.uber.com

77 pointspplonski8624 comments

I manage a ML team and the article has several ‘borrowable’ ideas, the primary idea supporting model builders working on their local Linux device and simplifying the process of transferring locally built models to a scalable platform. If we were starting today to implement our in house system I might just use Ubber’s system as a checklist for features to implement.

Not too off topic: after mostly building and training models on remote compute (both at work and for my at home learning projects), I recently bought a fast laptop for home use that has a 1070 GPU and the experience of using a local device with a GPU makes me regret many years of SSHing to remote compute or using Jupyter remotely for machine learning development. I mention this as a lesson learned that may be helpful.

The local dev painpoint is real. I was working on a library that can intercept and transparently offload keras fit() calls from ones local machine to a faster remote machine. Unfortunately, it seems most people just use GPU workstations and don’t need the speed improvements locally.

I recently bought a fast laptop for home use that has a 1070 GPU and the experience of using a local device with a GPU makes me regret many years of SSHing to remote compute

Most interesting problems in my experience are constrained by I/O bandwidth, and being able to run C/GPUs for extended periods without thermal issues. You really need something in a DC for that. I chose not to bother with a GPU laptop myself because there just seemed to be no point, for ML at least. I guess just for playing, it’s a nice luxury, but I can’t imagine relying on it for real work.

Gaming desktop PCs exist for a single reason - "being able to run C/GPUs for extended periods without thermal issues." If you were doing this for actual work, you'd obviously need to step upwards to a real workstation, which would actually be built quite similarly to a bog-standard gaming desktop save for using higher reliability components, e.g. FirePro- or Quadro-class GPU hardware. But to suggest that you need "something in a DC" to explore interesting problems seems quite wrong to me.

Uber's software doesn't exist in the public realm, and so there is nothing to learn from it. It is imaginary fluff. I have posted a list of the ones that do actually exist in another comment. And you should've at least got a 1080.

It actually gives a lot of idea and validation for concepts that I was thinking about for my project https://github.com/polyaxon/polyaxon, namely for going from trained model to serving those models on other platforms in a seamless way. I also think that high-level articles like this one are really good for people who are trying to build internal tools around ML/AI ops, people can get more or less ideas about what and how other companies are managing their ML pipelines and how they impact the productivity of their teams.

You should be getting ideas from actual open source projects that actually exist, not from imaginary/fake ones. There are numerous serving packages that actually exist and offer ideas.

I wanted a portable laptop and just saw 1070s on what was available. What brand of GPU laptop do you suggest?

I got some ideas from reading the article, but yes, I agree with you that the article should have a link to a public git repo for at least some of their tooling.

EDIT: I wanted the multi-GPU Lambda Labs box, but with my lifestyle I require some mobility. I agree that the 1070 is a bit weak, but is still better than no GPU and I like The System76 laptop I bought even when I am not using the GPU.

You will get a lot more ideas from other software that actually do exist. I can easily write a fake blog post with fake thoughts giving you fake ideas, but that's not what you want; you deserve ideas that work.

System76 does sell laptops with the 1080.

As it stands, this article strikes me more as a recruitment device than something useful to ML practitioners outside of Uber. I'm hopeful Uber will open-source parts of the Michelangelo stack, so it will be a contribution to the broader community.

Disagree. The article validates a lot of thought we have been putting in our own company. Not everything has to be open sourced to have value.

Pretty slick, but more impressive is their well written overview of the tool.

It has some interesting model def features: The model interfaces abstract away where and how the model will be deployed so that the user can focus on what they understand best: the data going in and out of the model.

Are you serious? There is no code for us, so there is no tool! The article is as good as fake news.

Not every company needs to open source their tools. Uber has contributed a substantial amount of open source tools. This product is clearly new and 99% likely ties to their stack meaning it is useless to you anyway.

Make your own tool and open source it if you are going to be so mean. I would never want to share my hard work with people who act this way.

Related meetup:

UBER : Big Data Infrastructure and Machine Learning Platform https://www.meetup.com/SF-Big-Analytics/events/257037259/

Anyone working at Uber that can highlight the non trivial differences between this pyML Michelangelo and Kubeflow? I am very interested in hearing first hand opinions about pros and cons

I wish my company was cool enough such that our internal tools (even if they're technically complex) would be of interest to people :(

Who cares about this garbage if the tool isn't even open source? There are lots of ML deployment tools that are open source. I know haters will downvote my post, but it's the truth. If I can't actually fork and evaluate a tool, it is hyped up garbage to me.

Meanwhile, here is a list of open source ML deployment packages:

https://github.com/oracle/graphpipe

https://github.com/eliorc/denzel

https://github.com/tensorflow/serving

https://github.com/ucbrise/clipper

https://github.com/DLHub-Argonne/dlhub_sdk

https://github.com/kubeflow/pipelines

FWIW, the "Guidelines" ask you to please not dismiss work out of hand. Appearances matter, too - you can point out that a tool is proprietary and has lots of plausible FLOSS alternatives without being uncivil about it ("garbage").

Edited: Commenting about the voting, especially in the way you did here, is inappropriate too. We're here for substantive discussion.

I agree with you, but I also think there’s room to point out that Uber has lost credibility and it wouldn’t be surprising or inconsistent if it was just a PR post for recruiting hype. Without details (not mere surface comments) on how it is differentiated from the many other available solutions and deep dives into what use cases it is specifically better suited for, it seems reasonable to treat it with a lot of skepticism.

But to be clear, I totally agree in terms of the degree and tone. I don’t think the parent comment you responded to was “uncivil” in any way, but overly dismissive instead of just noting to be skeptical.

Skepticism is one thing, lack of civility is quite another (and I'm sorry, but calling stuff "garbage" in such an off-hand way is clearly uncivil to me, especially when shallow dismissals are expressly pointed out as a problem). See brylie's comment in this thread for how to do it right - take that, add a stronger caveat about this tool's proprietary status, in contrast to so much other stuff in this space being FLOSS, and it would make your point quite well without degrading discussion.

I didn’t write a comment calling anything garbage. In what way did my comment degrade discussion?

I believe they’re referring to 'Michaelanjello’s comment upthread:

https://news.ycombinator.com/item?id=18692801

Well my comment itself was downvoted, so I don’t agree it was a focus on the original comment.

Reply on news.ycombinator.com