Week 1: AI using the tf.keras.Sequential model
Hello everyone!
This week was quite the introduction to my senior project. I started with researching on Monday and transitioned to programming on Tuesday. Then I ran into a roadblock, which was that one of the functions I was calling (tf.keras.utils.image_dataset_from_directory() specifically) expected a folder for each class, or make of the car in this case. The problem was that the dataset I found, which I will explain later how that works, had all of the images in one folder. If this were a small amount of images, I would have just manually organized them, but this was 64K images of cars and the filenames were so long that half of the name was just collapsed to “…”. So I had to find another solution. I ended up going with possibly one of the most unwise decisions. I chose to code another program that organizes the files for me. If I had done it manually it would have taken an hour maximum to organize the files, but because I wanted to code it, it took an hour and a half to code and fifteen minutes for the program to move the files to where they needed to be. Now back to where I got all my data from, I used a web scraper to scour the internet for pictures of cars. It took about 50 minutes on Saturday because I was preparing for this project and didn’t want to count this in my hours, but when I looked at the images, there were interiors in the dataset, which was 122k images. I was panicking when I saw that the website that gave me the web scraper also had a dataset of 64k images, so the crisis was averted.
Right now, I am at a point in the project where I’m about halfway done coding the first algorithm and have yet to run the whole thing. Surprisingly, the debugging hasn’t been that bad, which is a new experience for me, but that goes to show the change that proper research can make in the programming experience. I still don’t quite understand the purpose of every mathematical function that goes on behind the scenes, but that’s what the rest of the week is for. I understand a lot more of it than I thought I would though, which is an uplifting thought.
Throughout the last few days, I’ve also found out about how little I knew about cars. For example, before I didn’t know that horsepower and mileage are inversely proportional, but through studying the names of the images in my dataset, because the names are so detailed, I found a correlation. In the next few weeks, I’m sure my knowledge about them will blossom into something more useful in the real world.
As of right now, I’m planning to finish this algorithm this week and move on to the next algorithm next week. Overall this was an eventful week but also very fun.
Thank you for joining me again,
Akshith