My diary - Week 2 - Fast.ai lesson 2
Monday August 1 2022
Today I tried to work on my spider classifier but realized that the data collected on the internet wouldn’t work: a lot of them are mislabeled. I might still do this classifier later by using a Kaggle dataset. In the meantime, I will change categories. I’ve thought about clouds, crocodile/aligator (but I would have the same problem). I looked briefly into healthy food, legume/vegetable… Hamster/Guinea pig would work well but it’s not very interesting. I’ll think about something else.
In my other courses, I learned about map, lambda and list comprehension in Python. I need to work more on it.
Tuesday August 2 2022
I’m thinking about a new classifier to make. My idea for now is a model that would recognize different types of cartoon drawing style: Disney, Marvel, Ghibli. This would allow people to download their drawings and see what they are closest to. Maybe a 4th more realistic category could be nice. Maybe I could go with famous painters’ styles like Picasso, Van Gogh… This app could be fun. But I think I’ll stick with the cartoonish style.
In ML, I learned about linear regression and the cost function. I loved the representation of the cost function in different ways:
Wednesday August 3 2022
No progress on my project today: I want to spend time reading Deep Learning for practical coders before going back to it. I’m still going for a drawing style classifier. Sounds like something my family and friends could enjoy.
I spent some time studying Numpy and taking notes on a Colab notebook. Someone on Twitter was interested in seeing it so I’ll post it here. Jupiter Notebooks are perfect for taking notes on programming courses! And being able to, very easily, turn it into a blog post is the cherry on top!
In the ML course, I’ve started learning about gradient descent. I like how my courses crosses so the same topics are taught in different ways.
Thursday August 4 2022
Not a lot of time to work on classes today, but I finsihed reading ch.2 of Deep Learning for practical coders. Lots of interesting things about making a dataset and end to end production. I’ll try to work on my project tomorrow. I might start looking into Kaggle competition too.
Friday August 5 2022
I made some progress with my ‘drawing styles’ classifier. I chose 4 categories: Ghibli drawings, Marvel drawings, Disney drawings and chibi drawings. I made a dataset by downloading images from Duck Duck Go. I also removed the ones with broken links. I’m excited to keep working on it this WE. The next step is the DataLoaders!
I’ve also decided to nake meal plans and completely schedule my days. I might do a schedule for a week and review the priorities each morning. It worked well today and allowed me to do all the things that were important.