This sounds awesome! I look forward to seeing your units develop, especially “Ancient and Medieval African Civilizations”. I think that you’re absolutely right: African history hardly gets any mention at all in high school classrooms. I suppose that’s partially because of a relative dearth of written documents but it’s still inexcusable

Good luck with your work!
Tyler

This project looks extremely promising and your proposed curriculum seems wonderful! As an IR major I definitely agree that coursework on African culture, politics, and history can and should be emphasized better in the pre-college years. Many students do not have much of an introduction to these issues until they enroll in a course at school because of their own interest or (in some cases) to fulfill a GER.

I wish you the best and look forward to hearing about your project as it develops!

Sorry I let this sit for so long — I didn’t know anybody’d be interested in this work.

Almost any time you grow cells in culture, they’ll form some clumps. Most cells will keep growing until they get inhibitory factors from other cells they touch. This is a pain when you’re trying to automatically segment cells, because it’s hard to tell where the borders of cells lie. The current best solution (as far as I know) is to have a person sit down and draw out boundaries where they think individual cells are. This is time-consuming, error-prone, and non-reproducible, so I’d like to automate the process. The trouble is, most algorithms will identify a big clump of cells as a single cell, basically because they share a giant border. Identifying when it’s found a clump like that is the first step in correcting the segmentation algorithm.

So far, I’m just using the scripts to identify learning algorithms that are good at making good classifiers. There’s not much point in rewriting the algorithms — once I can verify that one works well, I’ll probably just keep using it. Right now, they’re working off of a bunch of statistics that the program CellProfiler takes by default — stuff like area, eccentricity, major and minor axis lengths, the works. Curiously, some of the decision trees use information like “mean x coordinate,” which shouldn’t have *anything* to do with whether or not they’re clumps. That’s a good indicator that at least some of the successful-looking algorithms are only spuriously successful.

I’m using Java because a) I’m much more familiar with Java, b) Java runs faster than Python (which usually won’t matter much, but some algorithms might take a long time), and c) I happened to find a good package of machine-learning algorithms for Java.

In general, it prompts the question of whether ensuring clean water supplies is inextricably linked to the development of power generation. The implications would be interesting to government and NGO clean water efforts in under-developed areas of the world such as Africa.