Capstone project first stages, and other thoughts

What can I say, it’s been a very busy last few weeks! I am now in the early stages of my final project, for which I applied to work with Digital Globe work on image segmentation. Code and presentation materials for my NLP project will be posted very soon as well.

This last week has been amoung the most frustrating of the entire program because I had to spend so much time wrestling with getting the Geospacial python packages installed in a working configuration. I had to learn the hard way that they do not play very nicely with Anaconda; so that ment creating a custom Linux Ubuntu AWS machine/Python environment without Anacaonda. As I have a background in windows server administration, not Linux, this was quite the painful learning experience for me, but early Saturday evening I am pleased to say I had all of the pieces up and running, and had an example U-Net learning from SpaceNet data on my AWS machine’s GPUs; and boy am I thrilled I can get back to data science and machine learning and put system administration aside for a while!

I don’t think I’ve posted yet, but I had been planning on using the SpaceNet dataset with OpenStreetMap data for ground-truth labels, to extract (by image segmentation) road surfaces, effectively generating an instant streetmap from the photos.

However Alan at Digital Globe keyed me into some recently released images of the Hustin TX area before and after the recent huricane, so I have tentitively switched focus to training a segmentation U-net to find water (hopefully, including muddy flood-water) and use it to label/segment flooded regions in the post-hurricane images. Wish me luck!

On a different topic, on Friday we had a special ‘technical interview workshop’ which I thought was extremely helpfuly, and included an hour or so of practice coding questions, which I thought were a lot of fun, I had most or all of a good solution to all but one of them, and apparently thought the whole interview process looks a lot less scarry than most of the rest of my cohort seems to. It could be that spending the last 7 years explaining my throught processes and answering techinical questions in front of my students has had some side-beneifits. Here’s hoping employers value good metacongnition and communicaiton skills.

Written on September 3, 2017