Day Zero Checklist

Part 1 (2018)
1

@jeremy if this isn’t an appropriate title or the post, feel free to move or remove :slight_smile: If folks want to edit and clarify any points below, I think that would be helpful for everyone to be in the best position possible before class on 10/30.

Day Zero Checklist

… or, what you should have ready to go by day 1 of class:

  • Bash/Terminal client

  • Git client

  • A development environment for machine learning (AWS, Paperspace, and/or your own rig)

  • AWS

    • Pros: What your employer is likely using, special treats (maybe), heavily documented.
    • Cons: More expensive, inferior GPUs (for now)

  • Paperspace

    • Pros: Easy to use, GUI or Terminal views, better GPUs than AWS, a fast.ai VM ready for use, $15 credits
    • Cons: Less flexible than AWS, more expensive in the long run than own machine

  • Crestle

    • Pros: Easiest to use, all fast.ai software installed and ready to use, can switch between cheap non-GPU and full GPU instance quickly
    • Cons: A bit slower than other options; no ability to ssh directly in (although can use web-based terminal)

  • Your own machine

    • Pros: Can be configured with superior hardware to cloud options, cheaper in long run.
    • Cons: Difficult to set up, more expensive upfront
8 Likes
2

Great! Thanks. I edited the title, since this isn’t actually a wiki (i.e. editable-by-all) thread, and added Crestle, plus made a few changes.

Most importantly, I removed the bit about cloning fastai repo locally. For most people, there’s no reason to do that (unless you have an Nvidia GPU of your own).

3

Ok thanks.

If you want to make it part of the wiki for folks to edit I think that would be nice; either way, I think lists like this go a long way in making day 1 of anything go smoother.

I’m planning to add some links when I get home and review my notes from the Saturday workshop.

4

Making it part of the wiki sounds great - feel free to copy it in there yourself :slight_smile:

5

@jeremy are we supposed to request a p2.xlarge instance before class starts? I remember that Amazon was taking up to 1 day when I did part1 v1.

6

No we’ll start using crestle for lesson 1. And in theory AWS is meant to be giving access to every id that was on the spreadsheet.

But it would be good to check, and request access if you don’t have it.

1 Like
7

My current limit for all P2 instances is zero.

image
image.png891×109 2.68 KB

Would p2.xlarge be right instance type that I should request for? I guess I can use instructions from the part1 v1 video.

For others:
Direct link to limit page if you are new to aws: https://console.aws.amazon.com/ec2/v2/home?region=us-east-1#Limits:

1 Like
8

I will suggest start with p2.xlarge. It come with one NVIDIA K80 GPU which is more than enough for learning IMO.

The next one (p2.8xlarge) has 8 GPUs which is an overkill. Also remember that price increases 7x as you go from 1 GPU to 8, and then 8 to 16.

–

1 Like
9

Thanks. I’ve requested p2.xlarge

10

Is there a script or a notebook to test if we installed everything according to the instructions?

11

I don’t think we’ve.