Overall Notes for the Day

Data Science

• started new 5000 batches with 40px tiles again now that randomization is right, tiles on way tolerance set to 25%, 10px (laptop run)

• data science seems to be at a plateau, need more programming

Programming

I borked a Linux box I have by hard restarting it. I ordered a Linux USB key to reflash it, so I’m blocked on that for a few days.

Since my GPU plans aren’t working out, and my predictions aren’t improving by messing with the parameters, I can do a couple of things, which I’ll work on in this order I guess:

• add more area/data to the analysis
• assign more pixels for roads, based on road type (this will turn on more tiles, should be more accurate)
• improve the neural net algorithm to something more appropriate for RGB-IR data
• automate a series of runs across my variables (patch size, tile size, has_ways algorithm for tile)