Platform: Kaggle Kernels

[wdhorton]

The fastai library v1 is now available on Kernels!

Edit: Kernels were last updated on 24/2/2019
Note: The kernels are being maintained by @init_27
If anyone would like to contribute to the kernels, please ping @wdhorton or @init_27.

中文版

这篇post的中文镜像版正在建设中，欢迎大家来分享和共建！

FAQ:

1. learn.save gives error

Kernels load datasets from ../input/ this is read only storage. If we create a learner from here and try learn.save() , the function tries to save in the same folder as the dataset. Hence the error.

Fix: Set the path in learn.save() accordingly.

2. I don’t see my kernel output files

To see the outputs, you need to “commit” the kernel, which will run the nb from end to end. Thus creating your output files.

3. Uploading custom datasets

You will need to create custom dataset by uploading it to kaggle datasets and then linking it to your kernel. Follow the kaggle kernels tutorial to understand this.

Hope these solve the pain points.
Please let me know if I’ve missed anything.

4. HTTP or Connection Errors

Ensure that Internet connection is enabled in your kernel, this might require you to complete filling your kaggle profile.
Once you’ve enabled it, you’ll have to restart your kernel

5. Bus errors

Try setting num_workers=0

Other Pitfalls:

• Ensure GPU is enabled on your kernel
• If you get strange errors/‘not defined errors’, its possible that the kernel hasn’t been updated to the latest, ping @init_27 for such a case.

Lesson-Wise Notebook Links:

• Lesson-1 Pets
• Lesson 2 Download
• Lesson 2 SGD
• Lesson 3 Camvid-tiramisu
• Lesson 3 Camvid
  Note: Known issue, notebook needs more fine-tuning since its slightly compute demanding
• Lesson 3 Head-Pose
• Lesson 3 Planet
  Note: Learn.export error is a known issue.
• Lesson 3 Tabular
• Lesson 4 Collab
• Lesson 4 Tabular
• Lesson 5 SGD-MNIST
• Lesson 6 Pets-more
• Rossmann data clean
• Lesson 6 Rossmann
• Lesson 7 Human-numbers
• Lesson 7 Resnet MNIST

These 3 notebooks still need to be fine-tuned due to the compute restraints from kaggle kernels.

• Lesson 7 Superras GAN
• Lesson 7 Superres Imagenet
• Lesson 7 WGAN

Notebooks by competition (Created by Community, these are not being maintained)

• WiDS Datathon 2019
• Histopathologic Cancer Detection
• Home Credit Default Risk
• Plant Seedlings Classification
• Human Protein Atlas Image Classification
• Santander Customer Transaction

[init_27]

Oh, awesome!
I’m working on the same!
@wdhorton if its okay with you, We could take the alternating lessons and split the work?

[wdhorton]

Sure! You want to take 2,4,6 and I’l continue with 3 and 5? There are a lot of notebooks for lesson 7 so maybe we’ll split those up as well.

[init_27]

Sure, Sounds good to me!
Thanks

[wgpubs]

Can we use python 3.7 with the available kernels?

[jcatanza]

All the notebooks from the Fastai Machine Learning Class are available on kaggle kernels:

Fast.ai Machine Learning lessons on Kaggle kernels

@jcatanza and i have made ML lesson available at Kaggle kernels. Just fork the notebook and have fun! Lesson 1 notebook Lesson 2 notebook Lesson 3 notebook Lesson 4 notebook Lesson 5 notebook Let us know if you have any feedback or comments.

[miwojc]

Possibly some of ML notebooks will need to be modified for fastai v1 or maybe there’s a way to downgrade to 0.7…

[miwojc]

(Note: we’re keeping the Kernels private until the release of the MOOC, at Jeremy’s request. If you want to take a look at them before then, I’m happy to update the permissions with your Kaggle username so you can view them.)

thanks! i would be interested to access the kernels, as my attempt to run what’s your pet notebook in kernels resulted in 'RuntimeError: DataLoader worker (pid 114) is killed by signal: Bus error.
my kaggle username is: miwojc. thanks again!

[wdhorton]

@miwojc just added you to view the kernel!

[Ajaxon6255]

Horton please add me to your kernel access list also I have a running lesson 1 notebook with the new fastai sw:
https://www.kaggle.com/allanjackson/fastai-lesson-1-new-sw it is set to private for now and I will add you to my kernel access list. I was originally having the “bus error” but you have to set up the lesson 1 kernel from scratch -feel free to fork.
Kaggle name is Ajaxon6255 and I used your old fastai 0.7 kernels many times - thanks!
I will add @miwojc and @init_27 also.

[rohitag13]

Hey! really interested in having a look at your kaggle kernels. Do add me.

[init_27]

@wdhorton I’m done with Lesson 2,4,6.

There are 6 Nb(s) in Lesson 7, I’ll take “lesson7-human-numbers, lesson7-resnet-mnist, lesson7-superres” if its okay?
You can take the other three?

I’ll add your kaggle id to my kernels as well.

[init_27]

@Ajaxon6255 My kaggle user is init27.
I’ve added you as well.

[wdhorton]

Sounds good!

[Bhuvana_ka]

@Ajaxon6255 @wdhorton @miwojc @init_27

Can you please add me as well. My Kaggle user name is bhuvanaka

Thanks

[martijnd]

Can you add me as well? My Kaggle username is etinuz,

[gerardo]

Can you add me?
This is my handle gman123

[martijnd]

Hello Allan, I am having the same issue with the bus error. How did you solve this error? I am using v1.

RuntimeError: DataLoader worker (pid 143) is killed by signal: Bus error.

[Ajaxon6255]

I am in the process of running my Fastdotai Lesson 1 (new sw) Kernel from start to finish now on Kaggle. It has not shown the bus error. Martijn, did you have the buss error when running on the 17th or 18th of this month?
Google cloud servers had some issues on those days, but it may be entirely unrelated to bus errors you saw. I will add you to my lesson 1 kernel access (view and edit) and encourage you to fork it and run it in your Kaggle console. You will note that I have added a couple lines to help you check the fastai version and the torch version (both should be greater than 1). @init_27 and @wdhorton have all the other lessons that have an extra few bits of code that stop the bus error in the kernels they now have. (Oddly the few people on Kaggle that I gave access to my kernel - don’t seem to have access anymore according to my console). Access to my kernel (and theirs) is not public and will not be until the MOOC is publicly released. If you still get the bus error with my lesson 1 kernel after forking/committing/running in Kaggle please contact me and we will try to insure a proper execution for you (and the many others that will be using the kernels soon).

[martijnd]

Thx for your quick response and adding me to your kernel, I made a copy.

I had the error yesterday, so unfortunately it had nothing to do with the Google Cloud servers. I am trying to get the notebook lesson 1 running with the pets, maybe is has something to do with the read-only rights on the …/input folder.
I am able to run your notebook (lesson 7) without the bus error. Although I have some issues with the size. BTW Kaggle is now running fastai v 1.39.

ValueError: Expected input batch_size (17) to match target batch_size (64).

Edit: I now have my Lesson 1 Pets notebook running.

Next step is to create a solution-file file for evaluating the submission. Has anyone already has some experience with that?