Thursday With D&Z

Going backward, I was up until after 11. I had started cooking some eggs, but they were not ready, hard-boiled, and peeled until 11:30. Next, having put on my PJs at 10, I crawled into bed, my mind drifted, and soon I boarded the dream bus and did not wake until sunrise Friday morning. I also wrote the postcard for Mom Wild and put it out, preventing the recent morning rush to get it out before the post arrives on Friday.

Also, I spent part of the late evening exploring how to split data sets into training and testing collections using routines supplied in Open-Source Python libraries. I was still looking for the original source of the Titanic data set, but I have located two copies that match. This dataset describes the victims of the Titanic wreck, including their identification and whether they survived. It makes a good practice set for Python coding of AI-based classifiers.

This then led me down a rabbit hole on testing and validating interesting software options. Still, as usual, the write-ups used too many computer terms (yes, this even happens to me), and I was having trouble decoding their use and application. My interest was piqued, but I am a harshly practical person when it comes to coding, and unless I can find a clear explanation, example code, and a compelling benefit case, I will soon disregard it.

But I am aware that these obscure processes are specifically helpful for the last-mile work. I have learned that it is easy to develop an 80% solution for a classification problem (simply copying some code found on the Internet), but achieving above 90% requires taking special actions to improve accuracy. And with the overfitting issue (where your AI model becomes overly reliant on a few data elements in the test set that are not present in the real dataset), it can easily become a game of thumping moles rather than genuine analysis.  I suspect that this hard-to-follow testing method would be one of the ways to reliably improve an AI model to high success rates. More studying required!

I was doing the Alice-in-Wonderland reading, as I have offered to give a talk on using the Titanic data set for the HOPE_16 conference. I am preparing a short talk on an introduction to Python and AI.  I’m not sure they will accept an introductory talk in such an over-the-top hacker audience, but I’m waiting to see.

Before this, and I regret not getting a photo, Dondrea, Z, and I shared a meal at The Cheesecake Factory. We sat in the bar area and had happy hour items followed by massive desserts. The food was good.

We mostly chatted about the American Civil War (ACW, gamers play both the ACW and the English Civil War, and thus, the clarity is required), as Z was starting that in class. Z asked what our favorite war was to study, an interesting question. I went with WW1 and Dondrea, Vietnam. Dondrea views Vietnam as a defining example of small wars with modern technology, a phenomenon now prevalent worldwide. I also view WWI in a technological sense. I perceive the start of WW1, its execution, and its aftermath as a lesson in the misunderstanding of technological change and political systems that deny the future and are then replaced (or, in the case of Russia, executed). For me the best examples of technology impacting the world are the rise of ironclads in the ACW (in a moment all navies were obsolete) at Hampton Roads 1862 and WW1 Jutland 1916 were bew technology failes to let the Royal Navy to decisivly defeat the inferiero German Imperial Navy in what was a clash of fleets–the only one in the war. We discussed Gettysburg in ACW and some aspects of Vietnam.

Moving further back, I picked up Subha at her house and drove to meet Scott at Cedar Hills McMenamins. Subha, I won’t tell her story for her. She was available for lunch, and Scott and I worked with her for years. It was great to have her join us for lunch. We caught up with her and she decided to try the turkey sandwich with soup and salad, the lunch box special at McMenamen. Our new go-to. She had coffee while Scott and I had beers.

Before all of this, I spent the morning doing the blog, getting ready for the day, and updating my transactions in Quicken. I had risen just after 6 as I had trouble sleeping. More pollen-driven discomfort, and it was warm yesterday (80F+), making sleep difficult. I was tired most of the day.

Thanks for reading!