General course chat

(Mehmet Alican Noyan) #428


Just started setting up the new course. When I use the promo code for paperspace it says " Error! Could not claim promo code (code is expired)". Just wanted to let you know.


(Cedric Chee) #429

Hi Forum Admins,

I noticed that the access to the “Advanced Discussion” sub-forum is still private even after the launch of the course to everyone. Is this intended?


(hwdong) #430

I found your site and spend a whole day ,It seems a great work!
But I don’t know how to start. How to set up ? The site and website seems very very messy for me.
I read and searched forum and watched video.
But the video is not clear for me. How to start?
It is very hard for beginners although I am familiar with deep learning and tools (caffe,TF, pytorch) for several years. Your course seems hard to follow.

(Abhimanyu) #432

You can start with Google Cloab. If you’re not sure how to work with Google Cloab. Do let me know.

(hwdong) #433

Thank you .
I can run lessons on Google Cloab

(Gaurav Kolekar) #434

@jeremy are we also getting a new machine learning course?

(xnet) #435

Are there special installation steps for running the fastai code in a windows machine?

(Trinath) #436

I am trying to run the untar_data() function on kaggle kernel, i am getting error:

ERROR NOTE: ConnectionError: HTTPSConnectionPool(host=‘’, port=443): Max retries exceeded with url: /fast-ai-imageclas/oxford-iiit-pet.tgz (Caused by NewConnectionError(’<urllib3.connection.VerifiedHTTPSConnection object at 0x7f029e361630>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution’,))

Can anyone let know why this happening?

(Dhyey Pankaj Mehta) #437

Hey @jeremy, I just created my own image classifier after seeing lesson 2 of the deep learning course. It was a classifier to distinguish between guitars and violins. I was so happy to get a 98% accuracy on my classifier. Thanks for helping me create this!!

(Brian Smith) #438

Pytorch 1.0 is officially released but I think if you search you will find others. Windows is running slower than Linux right now - threads on that too

(Avtar Khalsa) #439

I am trying to get ImageDataBunch.from_folder to work and I keep getting this warning:

UserWarning: There seems to be something wrong with your dataset, can't access these elements in self.train_ds: 10

I am calling it this way:

mini_bs = 16
jet_data = ImageDataBunch.from_folder(jet_path, ds_tfms=get_transforms(),
                                      valid_pct=0.25, size=224, bs=mini_bs)

Where jet_path is the directory with my data. Within jet_path I do not have a “train” or “valid” folder and instead just have jet_path/class1 containing all my images of class1, and jet_path/class2 containing all my images of class 2. I tried setting up the directory as jet_path/train/class1 and jet_path/train/class2 but that did not seem to help. Does anyone have any idea what I might be doing wrong?

(Bernd Helzer) #440

I am excited about this class. I was trying to create a sample application with my model.
I can just train it from a jupyter notebook and save the model. According to the documentation, I can use predict on vision.learner. My question is how would I do this for a production system, where my images come from a video.
I could use opencv or something like it to pull out images from the video, then load each image and call predict on it. (I used opencv to pull out images for my training set).
I that the best approach, or is there a better way of doing this? Also what would you recommend as a server? AWS?

(Akshay Bhardwaj) #441

The URLs.PETS you are trying to untar is not really available in the Kaggle environment/Docker. The URL is for data to be downloaded from Amazon servers (where the dataset has been hosted by fastai devs) which won’t be possible in Kaggle kernels. Otherwise, if you try untar_data() on a kaggle dataset itself, it works.

(Akshay Bhardwaj) #442

You probably will have to use OpenCV to pull out images and feed to the model in production as well (I did the same when I was playing around with videos).
For server side, I prefer GCP, but if you are familiar with AWS then you may try it as well.

(Damjan Vlastelica) #443

Hi there. Just started the course. Hardest part so far has been navigating these forums :slight_smile: Anyhow trying to post every day so hopefully get better at this. Course has been great so far and Jeremy’s teaching style is relaxed and efficient.


Hi all,

Nice to e-meet you. These forum is definitely not trivial. Up to this moment, I am not sure how to write a new message.

I would like to discuss installation for working with Google Cloud Datalab. I was able to run
pip install fastai
from fastai import *
from import *

The last command, fires an error:

ImportErrorTraceback (most recent call last) <ipython-input-14-c0e76450f370> in <module>() ----> 1 from import * ImportError: No module named vision

Please advice.

(Joseph Catanzarite) #445

TOPBOTS, run by Mariya Yao, offers weekly summaries of the latest AI papers organized by topic. Looks useful for keeping current!

(benjamin Dubreu) #446

you should probably post that question in the “fastai users” subsection of the forum. It’s for everything related to the library itself :wink:

Note that I would help you with the answer if I could, but I have no idea what’s going on.

(benjamin Dubreu) #447

Hello everyone !

Does anyone have a clue where it would be possible to stay in SF if I was to come for PART2 of the course ? I’ve looked at Airbnbs but they are either overpriced, or too far away from the Data Institute…

(Dhyey Pankaj Mehta) #448

how to create my own validation and test set and use them in an image classifier?