June 10, 2025
Day 11 – Starting the Code
What I Learned
Today, I started the day by watching a video on ensemble models (stacking). To understand one concept mentioned in this video, I had to watch another video on k-fold cross-validation. I also learned about leave-one-out cross-validation, in which each sample is treated like a whole block/fold and tested on the algorithms. The common practice to divide the data is ten-fold cross-validation. I also learned some utility functions in Keras, such as to_categorical and image_dataset_from_directory. The way we detect the GPU in TensorFlow is different from PyTorch, and I found the TensorFlow method a bit harder.
After watching everything, I started building an ensemble model to test with our dataset. I will be stacking MobileNetV3 and EfficientNetV2L at first. I still haven’t finished the code though, and will continue tomorrow.
Blockers
No issues faced.
Reflection
Today begins the fun part as I start coding on my own. One thing I don’t like is how complex TensorFlow is, yet how much less popular it is compared to PyTorch, which I have experience in and find easier. I’m also looking at some other codebases I found on GitHub to see if I can adapt any of their code into mine. I am excited to see where my code takes me in terms of performance and accuracy on our dataset.