Moved- See Edit! A Guided Walk-through of 2.0 (Like Practical Deep Learning for Coders)

Cool. Note that sometimes eps is inside the sqrt(), and sometimes outside, depending on the implementation. You should make sure that your 1e-6 is consistent with the location of the eps you’ve used elsewhere.

1 Like

Thanks for the hint! I’ll go check on that and see if it helps later today!

It looks like the issue with Colab has been fixed (they pushed 1.3.1). Double checking now. It’s good! I’ll post a headpose notebook shortly :slight_smile:

Posted!

Looks like that wasn’t it. All of ours have it outside (verified by @morgan)

it had left it outside the sqrt in both Ranger and RangerQH. RAdam in fastai v2 also has it outside and my port of QHAdam also leaves it outside. - morgan

Thanks so much for this guide!

It’s really useful to switch between fast AI v1 & v2, because I can get more insights into the fundamentals.

I got stuck at 02_Custom_Image_Classification when I use the cnn learner learn = cnn_learner(dbunch, resnet34, pretrained=True, metrics=error_rate)

The Learner object that is returned has no attribute ‘fit_one_cycle’. I used the help function on the learn variable and the fit_one_cycle method doesn’t show up.

I’m using Google Colab, I did restart the runtime and I get the same error. I’ve also rewritten the code & pasted ~approx everything you wrote in the guide and same error showing up.

n.b: my Leaner object also doesn’t have a lr_find() method.

2 Likes

@andreihrs Try doing:

from fastai2.callback.all import *

And thank you very much for the kind words :slight_smile: I’ll update the notebook with that fix later today.

3 Likes

I’m working on NLP next, I plan on trying to migrate Multi-FiT over once I finish with an IMDB sample and IMDB (so we can see from_csv and the original from folders). If anyone has particular topics they want me to do for notebooks please let me know :slight_smile: Otherwise here is the outline of what is left in no particular order:

08_Rossmann - Will wait until tabular issues are fixed
09a_IMDB_Sample
09b_IMDB_Full
09c_IMDB_SP/Forward/Backward/Ensembling
10a_MultiFiT
11_Segmentation
3c_Baysian
(Possibly GAN and/or object detection)
Multiple points (pose detection)
DeViSe

4 Likes

thanks much @muellerzr, an implementation of DeViSe in 2.0 will be helpful

Hi @harikrishnanrajeev, is this what you are talking about? (I have not heard of DeViSe until now) If not, could you point me in the right direction? :slight_smile:

yes, correct. This topic was also covered in Part 2 lesson 11 wiki. Class video is here

1 Like

Got it :slight_smile: I’ll see what I can do.

1 Like

@harikrishnanrajeev it was taught in the most recent part 2. here. I’ll do my own re-implementation but it’s pretty good

2 Likes

I’ve added in an Object Detection notebook based on RetinaNet in 2.0. This also shows how to do multi-output predictions for bounding boxes (until they are supported in the library)

6 Likes

Hey @muellerzr, were you able to get SOTA on ImageWoof? I tried running your notebook and tried replicating lessw2020, but no success…

1 Like

There are issues (like I mentioned in the notebook). We’re working on figuring it out slowly. We think it has something to do with the transforms. Know that all the implementations are correct. :slight_smile:

Also there are issues with the tabular notebook. I’m trying to get to it this week. (They updated the tabular API)

1 Like

Is there any specific thread where you are discussing what might be the problems there? I’m very interested in participating :smile:

See here: Meet Ranger - RAdam + Lookahead optimizer

@morgan has been working hard at cracking this :slight_smile:

1 Like

We had a problem in the cnn head that was fixed yesterday. So things might work better now.

2 Likes

Can confirm it did help a bit :slight_smile: Got about 72% on average of five runs, I’ll update the SOTA notebook too. Thanks Jeremy!

2 Likes

So a quick update to these, (I promise I’m not ignoring your requests!) I am setting up my lecture material for my study group. The new version will be live streamed so anyone can get involved and to my knowledge will be the first using fastai 2.0. I plan on starting in mid-January next year. I will be including all of the types of models mentioned here and walking through them. For those interested, keep an eye out here in the next week or so for an official mega-thread to discuss it :wink: (and if you have any questions feel free to DM me!) This new version for streaming will be new to me, but I’m hoping that it’ll go swell :slight_smile:

Edit: Also these current notebooks are outdated In terms of how to install the library, I’ll do my best to update them in the next week or two depending time. In the meantime, follow the install directions here: Fastai-v2 - read this before posting please! 😊

Edit x2: All notebooks are updated and Rossman is there too :slight_smile:

15 Likes

Hi muellerzr Hope your having a fun day!

Your output and support always inspire me.

Long may it continue mrfabulous1 :smiley::smiley:

1 Like