4 ms·
Ask HN: How are you doing machine learning, especially in the “cloud”?
Assuming you do it professionally or are involved in a team or organization that use machine learning, how do you go about your machine learning projects? What is your workflow? What tools do you use and why? What tools don't you use and why not? How did you evolve and come up withe way you're doing things?
How is your team organized if you have a team? How do you collaborate and hand-off each piece of the puzzle? How do you handle data, develop your models, track experiments, deploy and manage your models, etc.
Which cloud provider do you use? Do you use their "AI" offering, or do you use them purely for the infrastructure (compute, storage, containers, etc)?
What's onboarding a new team member like? What do they have to learn? Best practices you came up with or follow?
I like interacting in threads such as this one[0] or ones addressing frictions.
- [0]: https://news.ycombinator.com/item?id=25938654
- jimmyvalmer 5y agoBruh, your sh_t posts here and HN are among the worst offenders in the category of palming off commercial solicitation as academic inquiry. Some [other guy got an earful](https://www.reddit.com/r/MachineLearning/comments/p1xrbx/d_what_is_your_workflow_for_budget_compute_on/?utm_source=share&utm_medium=web2x&context=3 https://www.reddit.com/r/MachineLearning/comments/p1xrbx/d_w...) a few weeks ago.
- kowlo 5y agoRegardless, the submission has been voted up... The naive position would be to simply allow votes bury it.
- Jugurtha 5y agoI was going to ignore your comments history and treat this comment afresh, but I'll mention this instead of directly answering on topic because rudeness and trolling seem systematic in your comments. Seeing what others in the field are doing is useful, especially in a rapidly changing field with no clear ways of doing things. The gains warrant to keep up to date on best practices, or at least the "what-has-worked-for-us-is-...". This means frequently asking questions and having conversations with people who are active in the field helps adopt, discard, and tweak practices. There is a "semblant of a somewhat right way" to go about a software project, there even is tooling to manage that with a rather stable taxonomy (repo management, services, CI/CD, merge) and most developers would familiar with process and tooling at different organizations. There are millions of people writing issues, creating branches, submitting pull requests, checking tests, merging branches, automatically deploying and monitoring software. Millions doing it in a very similar way. We have not reached that state in machine learning yet. - [0]: https://news.ycombinator.com/threads?id=jimmyvalmer https://news.ycombinator.com/threads?id=jimmyvalmer