Fastai2 and new course now released

fastai
1

fastai v2 and the new course were released on August 21st. fastai v2 is not API-compatible with fastai v1 (it’s a from-scratch rewrite). It’s much easier to use, more powerful, and better documented than v1, and there’s even a book (624 pages!) about it. The book is also available for free as Jupyter notebooks. fastai v2 is documented here: http://docs.fast.ai/ .

To see updates to fastai v2, please see the following thread:

To get notified of changes to that thread, please selecting “Watching” from the menu at the bottom:

image
image1124×640 49.5 KB

fastai v2 only works with the 2020 version of the course. It won’t work with any previous version. If you’re currently working through one of the existing courses, keep going! :slight_smile: The basic concepts you’re learning will be just as useful for fastai v2. There is no 2020 version of part 2 of the course recorded yet, and we don’t have a date for when that might happen.

The 2020 version of the course includes material covering both machine learning and deep learning. So there won’t be a separate “Introduction to Machine Learning” (although the old one will still be available).

fastai v1 will continue to be available, and we’ll continue to provide bug fixes (and accept pull requests for it). To pin your fastai version to v1 (i.e., to avoid it upgrading automatically to v2), run the following command (assuming you use conda):

 echo 'fastai 1.*' >> $CONDA_PREFIX/conda-meta/pinned

Then, when you’re ready to upgrade to v2, remove the $CONDA_PREFIX/conda-meta/pinned file.

The github repo for fastai v1 will shortly be renamed to fastai/fastai1, and the repo for v2 will shortly be renamed from fastai/fastai2 to fastai/fastai.

If you’re interested in getting involved in fastai2 development, or just watching my live coding sessions (which I do most days), connect to our Discord server, which is where I stream my live coding, and there’s some real-time fastai2 development discussion:

All fastai development forum discussion is in fastai dev. The forums are the best place to ask questions.

62 Likes
3

Is this equivalent to part1 from the last year? (including Machine Learning stuff)

1 Like
4

Roughly equivalent, yes.

5 Likes
5

I have just started out with Part 1 of the course and I’m on Lesson 2. Would you recommend pausing for the 2020 version, or moving on?

I ask this question because if there are drastic changes in the APIs and workflow from fastai to fastai2, it might be beneficial to wait it out. From what I can gather, this is not the case, but it’d be great if someone can confirm this.

4 Likes
6

I’d suggest moving on.

15 Likes
7

One thing I have learned throughout my journey in fastai, is to pick a use case and do it exceptionally well. The version might matter, for sure; But applying the concepts and modules is barely done the same way., despite changes in syntax and modules denomination.
Can’t wait for Fastai2.

10 Likes
8

Hi - I assume the Part 1 will be a top-down approach, which is more like an intro to Deep Learning, like in 2018 and 2018. My question is: would any of the more ‘advanced’ stuff be covered anytime soon? For example, GANs, Transformers or Object Detectors? Or is there no plan in integrating those technologies to Fastai V2.

Thanks

1 Like
9

Is part2 also covered in the book or is it separate content? Thanks.

10

From what Jeremy has said, a mix of both. The course covers the first half of the book, the second part would be the rest, with potentially room for other topics like object detection, etc (Jeremy’s words).

In regards to technologies, GAN’s are in there, with some object detection. That being said, those areas (and transformers) are a good place for smart folks to integrate it in (as it’s open source, so more than just Jeremy and Sylvain, they can only do so much :wink: ) For instance we have a HuggingFace port made already (currently an external library)

(Do note: by technologies I do not mean the course itself. It’s unknown, besides the rest of fastbook)

7 Likes
11

Thanks for the exciting announcement! Is there a target date for the official release of the 2020 course / fastai v2?

12

No - it’ll be out when it’s ready. Shouldn’t be too long now.

15 Likes
14

Please don’t do that. (IE @ing admins when there are plenty of other people who have taken the course live and are on these forums). This question has been answered on the other course-v4 thread. FastAI course v4 MOOC - #8 by muellerzr

1 Like
15

Thank you Jeremy & all admins for the hard work :pray:
The course, the book, the forums, now the Discord chat :+1:
Can’t wait for part 2 :crossed_fingers:

8 Likes
16

You wrote:

fastai v2 only works with the upcoming 2020 version of the course. It won’t work with any previous version.

I am struggeling a bit with your approach: I found your great video tutorials and repository some weeks ago. It was very easy for me to install the recent version fastai-1.0.61 on a Windows computer and run the notebooks from the “example” folder. Very nice. I like your approach to make coding available to everyone as quickly and easily as possible.

But: I tried to get into the notebooks from your “courses” folder, but these require fastai-0.7. Unfortunately , these require fastai-0.7 and after some research here and there I found that lots of people are struggeling with getting this older fastai-0.7 installed and running, because fastai-1.0.61 is already available. Thanks for leaving the short message [00-DO-NOT-USE-WITH-FASTAI-1.0.x.txt] in the “courses” folder.

Now you are saying, that with fastai-2 the same will happen again:

fastai v2 only works with the upcoming 2020 version of the course. It won’t work with any previous version.

I really find this a bit frustrating, because drifting deeper and deeper in installation routines is exaclty what I don´t like. So sad that the notebooks in the “course” folder don’t work and are useless for me, because they require fastai-0.7. So sad, that the current notebooks seem not to work in near future once fastai-2 is out there.

Why aren’t these notebooks “backward” compatible? And if a company decides to use your fastai-library for production: would they have to change also their current notebooks each time when there is a new fastai-release?

1 Like
17

Its true, you’ll have to install the new library version if you want to use it. But that doesn’t mean you cant use the older version. Only the new course notebooks require fastaiv2. If you want to run older course notebooks, you can run them using v1. The same applies to companies which have developed code using an older version (including v1). That code will still work. fastaiv1 will be shifted to another repository and fastaiv2 will now be known as fastai! Only if they want to use the functionalities of v2, would they require to rewrite their code!

18

But how can I get these older notebooks from the “courses” folder running? I already have fastai-1.0.61 installed on my own Windows computer, which was very easy (I have Anaconda and Jupyter installed already). But getting the older fastai-0.7 installed on my Windows computer seems not to be easy! With this how-to I run into errors. And also further research on how to install an older fastai-0.7 version on my computer were frightening because of lots of users, who didn’t manage it either.

19

There are many ways to do this, and many threads that discuss this. The most efficient way i know of is to create a separate conda environment for installing v0.7(that way it wont clash with v1). Though yes, its very cumbersome and frustrating.

1 Like
20

I am not sure if I want to follow this path on separating different fastai-releases in different environments. And I see the same problem approaching with fastai-2. :frowning_face: I am not sure if “fastai” is the right choice for me as these installation routines take a lot of time.

21

The version might matter, for sure; But applying the concepts and modules is barely done the same way., despite changes in syntax and modules denomination.
assume the Part 1 will be a top-down approach, which is more like an intro to Deep Learning,

1 Like