I’ve been out of coding practice for ages, but what I saw for my short stint on free Crestle was everything I’ve wanted to work with since the University of Washington in the late-1980’s/early-1990’s wishing I could have majored in cyborgs in addition to designing cutting-edge music software amid our thriving local Seattle music scene. Let’s just say my new Alienware 15 Nvidia GeForce with maxed out memory and CPU is in the mail (with payments to follow me to the grave…). If I chose wrong, please tell me now.
Initial instincts were to make my Cats and Dogs project about Nissl bodies and Futurama brain slugs to help promote the BRAIN Initiative. Lately I’ve been pushing Deep Learning biology on Twitter as if the future of humanity depended upon it. I composed a song where two parts overlap causing a fresh new beat I’d like to pick apart like conversations in a room. Satellite imagery and street views are also of interest. Personally, I’d like eyewear that points out who’s on Santa’s Naughty List,
What I’m hoping to discover from the fast.ai community are guides to best practices. I mean, this will be my first laptop computer, so that’s going to be a learning curve in and of itself. I tried some cheap notepad thingy years ago, but the system would click buttons I barely hovered over, so I didn’t use it. I kind of have a slight idea of what to do first when the Alienware arrives regarding updating antivirus and and OS first, but paramount is getting fast.ai set up properly and working in the three-day idle period after delivery, and all this GitHub stuff is new to me, even though I’ve had a project on there since ages ago that eventually became obsolescent with the advent of JavaScript’s Map() object.
I saw that there are directory tree patterns to abide by for fast.ai development. How about best practices for naming my new system? For instance, Anaconda installation balked that my Windows username contained a space (in the directory name). After setting up a secondary admin account to install Anaconda, running Python pestered for writing privileges, and mklink
failed. Having to explicitly run the box as Admin every time would be an unbearable workaround. Granted, that was not on a fast.ai compatible system.
Several years ago while watching the new Stanford lectures on stochastic gradient descent, I was aghast at how random their guesswork was in determining the the proper step size. I still am after watching the fast.ai guestimate where best to set the value by eye rather than run some tests to find the most optimal slope (for a given situation). It’s not like we have to wait 10 minutes to test it nowadays…and by “we” I mean “me too” as soon as I get up and running to join the club.
I’m also looking forward to the prospect of working with Augmented Reality so that I don’t accidentally run over Ned Flanders on one of his infamous Fog Walks, if you’re familiar with The Simpsons. There’s just so much I want to try with the fast.ai software engine, and I’ve only barely scratched the surface of the lecture series.
You really don’t know how giddy I am encountering your 2018 videos late at night the other day. I describe it as feeling like I just got to test drive an exotic sports car (without the worry of crashing it), and compare fast.ai to Backyard Brains in that they both bring university graduate technology to any science classroom. https://bit.ly/2ttDi8e
So is there a web page or ten I should read along the lines of, “So you have a new $4K electronic baby to nuture”? What do I teach her? What don’t I teach her? I mean, it’s one thing to create a cyborg; it’s another to teach it not to be a jerk.