(8.15am) In a Linux script, to run two processes, wait for them to finish before starting to do something else:
./create_train_data.sh train -ldm &How could I not have known this earlier? Still, better late than never. This will cut the time needed to do some data mining work I am doing in half; I should be able to check the results of the trained algorithm after lunch instead of waiting till tomorrow.
./create_train_data.sh valid -ldm &
(9.02am) Austromigration is when birds migrate within the Southern Hemisphere. There are birds in South America, that migrate within the continent i.e., they fly North for the winter!
(9.20am) One way to track the location of small birds is to attach sensors to them. GPS chips are too big and too expensive, so one device works by simply recording the time of sunrise and sunset. How is this enough? The time in between, solar noon, tells you the longitude. The length of day can tell you what the latitude is, although the accuracy of the latitude varies through the year (less accurate around an equinox).
(9.45am) Cornell University and the Audubon society run a citizen-bird-reporting app called ebirds which is similar to the mPING application that we created. Birders love the ebirds app because it allows them to keep personal records, but Cornell can QC the data and provide it for scientific studies. Might be a way to validate bird density estimates from radar ...
(10.35am) People take drills way too seriously. We had an "active shooter" drill at work. This is better than a fire drill because you can simply shut the door of your office and keep working. No big deal. But then, the drill was followed by a "hot wash" where the police officer said he was impressed that the place became a ghost town. Fine. Can we get back to work now? Nope, some one had a question. And more questions. People!
(11.00am) Google Reader is being decommissioned in July. Oh no! How am I going to keep up with my RSS feeds?
(1.15pm) Just because I think the ground clutter case was from KGYX doesn't mean it isn't really from KRLX. It might be better to more careful about such things before wasting 2 hours on debugging something that is actually working correctly.
(2.00pm) On the other hand, the process of debugging a non-problem actually led to several neat ideas that do improve the clustering algorithm (once I try it on the actual problem case). So maybe it's a good thing to be sloppy once in a while. I still have to figure out why it is not working though.
(2.40pm) If a few storms are embedded in clutter, using a height-based attribute and expanding to cover the whole cluster will lead to the NN being poorly trained. Said clutter will never be removed. (sorry)
(4.00pm) When smoothing per-capita county data, a population-density-based filter size might work better.
(4.20pm) NDVI from MODIS before and after severe weather events is a good way to determine damage paths of weather events.
(10.30pm) Feedly is a worthy replacement for Google Reader. Although I am not a fan of the wide open spaces, I will probably get used to it.
(11.00 pm) 2NT showing the minors is a wonderful preemptive overcall of a strong NT. (sorry)
(11.55pm) Jim Crow laws were actually a major economic drain on the south. "A total of four restrooms had to be constructed and maintained at significant expense in any public establishment that bothered to provide any for colored people: one for white men, one for white women, one for colored men and one for colored women." That is from Isabel Wilkerson's "The Warmth of Other Suns".