Fast.ai:Practical Deep Learning for Coders, v3

 4 years ago
source link: https://www.tuicool.com/articles/NBN3uen
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.

Looking for the older 2018 courses?: This site covers the new 2019 deep learning course. The 2018 courses have been moved to:course18.fast.ai. Note that the 2019 edition of part 2 ( Cutting Edge Deep Learning ) is not yet available, so you’ll need to use the 2018 course for now (the 2019 edition will be available in June 2019).

Getting started

Welcome! If you’re new to all this deep learning stuff, then don’t worry—we’ll take you through it all step by step. We do however assume that you’ve been coding for at least a year, and also that (if you haven’t used Python before) you’ll be putting in the extra time to learn whatever Python you need as you go. (For learning Python, we have a list of python learning resources available.)

You might be surprised by what you don’t need to become a top deep learning practitioner. You need one year of coding experience, a GPU and appropriate software (see below), and that’s it. You don’t need much data, you don’t need university-level math, and you don’t need a giant data center. For more on this, see our article: What you need to do deep learning .

The easiest way to get started is to just start watching the first video right now! On the sidebar just click “Lessons” and then click on lesson 1, and you’ll be on your way. If you want an overview of the topics that are covered in the course, have a look atthis article.

Using a GPU

To do nearly everything in this course, you’ll need access to a computer with an NVIDIA GPU (unfortunately other brands of GPU are not fully supported by the main deep learning libraries). However, we don’t recommend you buy one; in fact, even if you already have one, we don’t suggest you use it just yet! Setting up a computer takes time and energy, and you want all your energy to focus on deep learning right now. Therefore, we instead suggest you rent access to a computer that already has everything you need preinstalled and ready to go. Costs can be as little as US$0.25 per hour while you’re using it.

The most important thing to remember: when you’re done, shut down your server . You will be renting a distant computer, not running something on your own. It’s not enough to close your browser or turn off your own PC, those will merely sever the connection between your device and this distant server, not shut down the thing for which you’re paying. You have to shut this server down using the methods described in the guides below. Otherwise, you’ll be charged for all the time it runs and get surprised with a nasty bill!

Here are some great choices of platforms. Click the link for more information on each, and setup instructions. Currently, our recommendations are (see below for details):

  • If you’ve used a command line before: Google Compute Platform, because they provide $300 free credit, and have everything pre-installed for you
  • If you want to avoid the command-line, try Salamander, which works great and don’t cost much
  • If you don’t have a credit card to sign up for the above services, use Colab, which is free, but has a few minor rough edges and incompatibilities.

Ready to run: “One-click” Jupyter

These are the easiest to use; they’ve got all the software, data, and lessons preinstalled for you. They’re a little less flexible than “full servers” (below), but are the simplest way to get started.

  • Salamander (instant approval; no installation required; includes full terminal access; $0.38 an hour; $75 free credit for students)
  • Colab; (instant approval, requires minimal installation, free)
  • Paperspace Gradient; (instant approval, no installation required, $0.59 an hour; $10 free credit)
  • SageMaker; (requires wait for approval, not quite “one click”… but pretty close, $1.26 an hour + storage)
  • Kaggle Kernels; (Instant Launch, No setup required, Free, not always up to date and not as well supported by fast.ai)
  • Floydhub; (instant approval, no installation required, $1.20/hour + $9.00/month (100GB storage), 2 hours free credit)
  • Crestle; (instant approval, no installation required, $0.78 an hour (CPU), $2.40 an hour (GPU)

Ready to run: Full servers

Some installation required

We also have instructions for using these platforms, but they don’t have everything preinstalled yet:

For those starting out, we highly recommend a Jupyter Notebooks platform (Option 1)

  • Notebooks are the easiest way to start writing python code and experimenting with deep learning.
  • Renting a Cloud Server (Option 2) requires environment configuration and setup.
  • Building a PC requires environment setup and more up-front money.

(When we release Part 2 of the course, we will go into more specific details and benefits on both building a PC and renting a server.)

Jupyter notebook

Once you’ve finished the steps in one of the guides above, you’ll be presented with a screen like this.


This is the jupyter notebook environment, where you’ll be doing nearly all your work in the course, so you’ll want to get very familiar with it! You’ll be learning a bit about it during the course, but you should probably spend a moment to try out the notebook tutorial.

Your first task, then, is to open this notebook tutorial! To do so, click nbs and then dl1 in jupyter, where you’ll then see all the lesson notebooks. First, tick the little box on the left of 00_notebook_tutorial.ipynb then click duplicate.


You want to avoid modifying the original course notebooks as you will get conflicts when you try to update this folder with GitHub (the place where the course is hosted). But we also want you to try a lot of variations of what is shown in class, which is why we encourage you to use duplicates of the course notebooks.

Launch your copy of 00_notebook_tutorial.ipynb and follow the instructions!

When you’re done, remember to shut down your server .

Our forums

Got stuck? Want to know more about some topic? Your first port of call should beforums.fast.ai. There are thousands of students and practitioners asking and answering questions there. That means that it’s likely your question has already been answered! So click the little magnifying glass in the top right there, and search for the information you need; for instance, if you have some error message, paste a bit of it into the search box.

The forum software we use is called Discourse . When you first join, it will show you some tips and tricks. There is also this handy walk-thru provided by another Discourse forum (not affiliated with fast.ai).

PyTorch and fastai

We teach how to train PyTorch models using thefastai library. These two pieces of software are deeply connected—you can’t become really proficient at using fastai if you don’t know PyTorch well, too. Therefore, you will often need to refer to the PyTorch docs . And you may also want to check out the PyTorch forums (which also happen to use Discourse).

Of course, to discuss fastai, you can use our forums, and be sure to look through thefastai docs too.

Don’t worry if you’re just starting out—little, if any, of those docs and forum threads will make any sense to you just now. But come back in a couple of weeks and you might be surprised by how useful you find them…

About Joyk

Aggregate valuable and interesting links.
Joyk means Joy of geeK