There shouldn’t be ideally be any difference running in google-colab and VScode.
As mike pointed out the it’s probably because you are running in a CPU. On checking your code, I noticed the only error is keyboard interrupt which have done.
1 Like
I thought the same.
I just didn’t know how to switch to my GPU on VScode. I searched online, and got this NVidia Nsight edition, but that didn’t seem to help
You’ll need to have a GPU installed on the machine. VSCode is just an editor. It runs code against your base system. If your base system which your VScode is talking to, doesn’t have a GPU, no amount of downloads will make it work. It might be helpful to mention what your base system is like so people an offer suggestions based on that information.
2 Likes
Not sure if this piece of info helps in your debugging, but i tried your code on my desktop using VScode (with old gpu nvidia 1070) , it runs.
Got to experience the whole “DL is a black box thing” today. I was going through the image classifier to recognize dogs and cats and uploaded an image of a hyena and got back a probability of it being a cat equal to 0.997764.
1 Like
Just to narrow the surface area of your issue, can you try it from the command line?, i.e. without VSCode
Interesting. Thanks for sharing. So the next experiment might be adding lots of hyenas to your dog class, or alternatively adding hyena as a separate category. I am interested to hear how it goes.
created this 1 page summary for a study group
Video: FAST AI: How to build a deep learning model for your own data? - YouTube
9 Likes
I really enjoyed Lesson 1 … WOW !!! Machine Learning has come a Long Way.
So, as I ponder potential use-cases, as an Urban Farmer, I love growing tomatoes, but one of the things I spend a lot of time thinking about is aphid infestations.
I’ve had a lot of people ask, “Do I have an aphid infestation?”, I always say, I would love to look at the situation. Maybe a photo is exchanged or not, but how can I build a business around answering similar questions for home gardeners?
How about an app that can analyze images, like I would and tell if you have an aphid infestation or not?
I’m new to machine learning but in just an few hours, getting around a particularly nasty bug with the help of this awesome community, I just created a model that can indeed look at a photo and render an opinion on whether you have an aphid infestation in your tomato patch or not ? What? Mind-Blown!
This is the power of Fast.AI
I look forward to many more AHA! moments and a great start on a career as a Deep Learning Consultant or ML Engineer!
Either title sounds Amazing!
8 Likes
@bitcoindoctor amazing story! did you upload your model to HuggingFace spaces by chance? And what dataset did you use to train your model? Happy trails!
1 Like
I am so new, I’m just starting chapter 2. But, I will upload it to HuggingFace spaces.
I trained on “max_images=200” healthy tomato plants and infested with aphids images from duckduckgo and it’s remarkably accurate for (256, 256) images. I would want to spend the resources to go to (1024, 1024) resolution images for published models.
The training and published work from this course is going to get me a $100K+ job.
5 Likes
Here is my first blog post explaining the code of the DuckDuckGo Image Scraper from Lesson 1. Hopefully someone finds this useful.
5 Likes
Fantastic first post, @hsaldera ! Looks like you spent some time and got the fastpages blog set up for this 
- that’s awesome.
The only one suggestion that I have is generally I know “what” the code is doing, but I can’t understand “why” it is coded that way- explaining this would make it much easier for the readers to follow along the code.
For example, how did Jeremey find out exactly what url to call?
Well, turns out you can open duckduckgo on your browser, inspect the network calls and see what happens when you do an image search. It will fire a request to the same i.js in the code. So now knowing this, I can work my way around to doing the same thing for a new search engine by myself if I want to! 
Kudos again, and hope to see more from you on DebugLand!
4 Likes
Hi, I hope this is the right place to post this question.
I just started the course, watched the first lecture video and am reading the book. I chose to read it on Colab so I could run the cells. I am running the first cell that trains the model to distinguish cats and dogs (I ran the two cells at the very top of the document before this one). It is the cell under the “Running your first notebook” section. At the rate Colab is running the cell, it will take an hour to train. The book says it should only take two minutes. I have seen mention of GPUs but I don’t see any option in Colab to make them use GPUs. Is this normal? It has been running for 30 minutes and is about halfway done.
Thank you for any advice!
3 Likes
Hey @toasty! Welcome. You will need to make sure you’re running a GPU “runtime” in colab. This is just a fancy way of saying that you’ve selected a GPU to run those code blocks.
In the main colab toolbar / menu click Runtime then Change runtime type
When the popup opens, in the Hardware Accelerator dropdown, choose GPU
Now re-run your notebook.
You should be good to go! Let us know how you get on!
4 Likes
Hi, I am trying to understand how the search_images function work. But google has not been useful so far. For example, it is not clear what urlread does. I have looked into documentation but there are no examples. I am not sure why q is provided in the data dictionary. Then there is re.search. Is there any explanation somewhere about search_images?
2 Likes
Welcome @durraniu ! So glad you’re here
@hsaldera contributed a post that goes into this:
This should begin to give you the intuition and mechanics
6 Likes
Thank you!
I could not run full cells in Lesson 1 on Jupyter Notebook. I have always been stuck here. Any solution please
1 Like
Use WSL if you’re on Windows. (Or better still, use a cloud provider such as Kaggle, Colab, or Paperspace)
1 Like