/Killara

Various Data/ML projects I do for fun.

Primary LanguageJupyter Notebook

Killara

Repo for various things.

Deep Learning for Satellite Image Classification

This is a Manning Live Project that uses Semantic Segmentation to detect the outlines of surface water in satellite images.

It covers Geo related topics (SRS, data collection) to deep learning combining U-Net (biomedical image segmentation convolution network) with ResNet. Once trained (with limited data set), the model could detect surface water outline with 80-90% accuracy.

(Tensorflow/Keras)

3D Medical Image Analysis with PyTorch

Manning Live Project that uses PyTorch to emulate medical image analysis. This project reads neuroimage files in nib format, trains a Neural Net using pre/post transformation pairs. The goal is to generate post transformation image where one is lacking.

This LiveProject is decent, but a bit thin, especially for someone who already know deep learning and have experiences with Keras. I wouldn't this LiveProject to get into PyTorch (which is what I was hoping to do).

(PyTorch)

DataApp with Streamlit

Streamlit is a pretty good tool if you want to provide interactive web dashboard. For non-technical audiences, it's heck of a lot better than Jupyter Notebook, and it's a whole lot easier (for us devs) than using tools like Tableau. This sample project shows how to interactively visualize t-SNE based on perplexity. Spoiler alert, it just displays pre-computed data, not calculate on the fly.

(Streamlit)

Machine Learning Mastery Code Bits

Code written while studying Jason Brownlee's excellent Machine Learning Mastery materials.

Killara

Killara is a suburb of Sydney where I lived decades ago. Killara is an aboriginal word that means permanent or always there. I lived on Wattle Street, a few houses from Killara Park where I used to go watch the locals play cricket. The park is still there and they still play cricket.