Today a short but fun exercise. I exported a model that I taught yesterday, and embedded it into a simple Django app. Now I can upload an image using a web form, and I get a result informing me if there is an passenger plane, a drone, a rocket or a fighter jet on that image. All in very little code. Impressive.
So now I have a model that quite well recognizes different types of flying gizmos. One problem I have is that in my local Jupyter Lab the widgets do not work and so I cannot do an ImageClassifierCleaner. Something to debug.
My plan now is to embed the model into a Django app, because well, I like Django and I have an idea for a project.
I will be tagging those posts with the #stfnML hashtag so that there is an easy way to find them all.
#machinelearning 2/2
I'm almost at the end of the second chapter of the fast.ai book/course on Machine Learning. This chapter was mostly a reiteration and extension of the first one. The goal was to create a model to recognize different species of bears and to export that model to be used in a separate application. I did not go this way because it required to set up a Microsoft Azure account, no thank you, so I downloaded a dataset of flying machines from https://www.kaggle.com/datasets/eabdul/flying-vehicles/data
I'm doing again the 1st chapter of the fast.ai machine-learning course, this time from the book, not the video (I think it's much better this way), and I set up a Jupyter Notebook instance locally with GPU support. It's fascinating yet terrifying to see my low-mid-range GPU (RTX2060) churn smoke and fire when training the models.
Make you think, if small training models use so much GPU resources, how much power is spent on real-life trainings?
Here's the link to the course: https://course.fast.ai/Lessons/lesson1.html. It's free, and also you can run Jupyter Notebooks for free on the web.
I also setup and run the Notebook on my computer locally, and it was cool to see my GPU actually do some machine learning goodness.
I'm more used to writing Python in files and in REPL, but after some time I started seeing the benefits on Jupyter in this application. #stfnML
I finished the first chapter of the fast.ai course on Machine Learning, and I am amazed! It's incredible how easy now people without a technical background (aka me) can start diving into ML and actually doing something cool, like creating an image recognition script. By following the Jupyter Notebook code in the course I created a script to differentiate between images of birds and forests, and later on I modified it to recognize Airbus A320 vs Boeing 737 :D
stfn 
