Study Group for Pune


(Nachiket Tanksale) #5

Interested


(Tabish Shaikh) #6

Nice initiative for Pune city. Go Pune!!!


(Rahul Pathak) #7

Most of you would have done earlier fast.ai courses but if you are doing this for this first time, you will need to prepare the environment and install python and packages. Please let us know here if you need any help with that.


List of part 1 study groups
(Shreyas Singh) #8

How do I get started with it? Is cloud service recommended? Among them what is preferred?


(Rohit Patil) #9

Great! :slight_smile:


(Rohit Patil) #10

https://forums.fast.ai/t/setup-local-aws-and-faq/25298
This will help you.


(Rahul Pathak) #11

Preference is always local if you have a decent machine with atleast 7 GB GPU memory otherwise you will endup waiting a lot for training to finish. Otherwise go for cloud service and there are pretty neat and easy to use cloud services. But you have to pay for that. Check the above link shared by @rohitpatil.

Last time in Part1 2017 we received AWS credit, lets wait if something similar happens now too.
You can also checkout “https://lab.snark.ai/”, “https://www.crestle.com/” “Google Colab”, “paperspace” too.


(Nikhil) #12

I personally prefer, Google cloud solution. With pre-built image, I was able to deploy a new server in 5 min.
Also they have 300$ discount for a year.

Checkout this post for installation, https://blog.kovalevskyi.com/google-compute-engine-now-has-images-with-pytorch-1-0-0-and-fastai-1-0-2-57c49efd74bb


(Tabish Shaikh) #13

With the free credit may be i will try google cloud solution, lets see how it pan’s out :slight_smile:


(Bharadwaj Krishnan) #14

i am in


(Avish Patidar) #15

hey guys…you can also use https://medium.com/@prakash_31206/fastest-way-to-setup-fast-ai-course-notebooks-for-free-using-google-colab-gpu-and-clouderizer-c8a004e1d50d
with this, you can use google colab resource for free


(Nikhil) #16

I haven’t used Google colab in a while. But, I had issue with getting GPU for model. It seems your request will go in queue and execute only when GPU was available.

Is it still valid?

I would say you can use Kaggle Kernel if you want to use free GPU. But the performance is not comparable with dedicated server + GPU.


(Avish Patidar) #17

I used colab with the help of clouderizer few days back and was getting around 11gb GPU memory. I used it for small task like going through notebooks and tweaking parameters. But it cuts connection if the notebook is processing for a long time.
So if you want to train model then definitely dedicated server and GPU is always better.


(Saqib Shamsi) #18

Hi! I’m interested too.


(pradeep Kumar Mahato) #19

Interested


(Nikhil) #20

Hi guys shall we meetup this weekend to know each other n to discuss any thing in universe?. I have created a group for the same.


(pradeep Kumar Mahato) #21

Hi, anyone getting this error in lesson 1-pets notebook
from pathlib import Path
path = Path(’/storage/pradeep/data/oxford_iiit_pet/’)
path_anno = path/‘annotations’
path_img = path/‘images’
fnames = get_image_files(path_img)
np.random.seed(2)
pat = r’/([^/]+)_\d+.jpg$
data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224)
data.normalize(imagenet_stats)
Error:
ValueError: padding_mode needs to be ‘zeros’ or ‘border’, but got reflection


(Nachiket Tanksale) #22

Checkout your pytorch version. This link might be helpful - Errors and corrections in 005_dogs_cats.ipynb


(pradeep Kumar Mahato) #23

Yes, updating the pytorch version solved the issue.


(Nitin Patil) #24

Has anyone of you built a deep learning box. I want to build one. Please share your experience.