Tweets
Replying to @crulge
i mean there was nothing left of me to pulverize after november 2016, so, yeah, i think i’m done handling things.
(original)
Replying to @josecastillo
scratch that: three beers. https://mobile.twitter.com/josecastillo/status/1469148751252049920
(original)
it does feel weird, after a few short decades on this earth, to feel confident piloting this body through the abyss.
(original)
Replying to @josecastillo
Day 9: two beers at Sunny’s, and the first live music in a long time.
(original)
Replying to @miles5z
Crowd Supply, soon! Sign up there for updates :) https://www.crowdsupply.com/oddly-specific-objects/sensor-watch
(original)
Replying to @cabe_bedlam
I’m aiming for something that handles both centibeats and martian seconds! the thing is, it violates some early assumptions that are currently coded into the API for Movement, the watch face app. IDK. It opens a can of worms, but if it works, it’ll be worth it.
(original)
Replying to @cabe_bedlam
it is! in fact it’s 255 256ths of 8 :)
(original)
Replying to @josecastillo
(yes, this gadget and this number are relevant to the code I’m writing)
(original)
Replying to @josecastillo
total side note: i got one of these calculators with the intent of taking it apart and hacking on it like the watch, but as soon as I put it back together, I realized it’s just an indispensable tool as-is. It’s amazing: no battery; just works. Purpose-built gadgets are wonderful.
(original)
Replying to @josecastillo
update: i am writing the code.
(original)
I’ve had this bit of code stuck in my head for days, but there’s so much work to be done with logistics and spreadsheets and words. All of this work is necessary and valid. But yeah, this code. It’s like a song stuck in my head and it desperately wants out.
(original)
Replying to @tahnok
I sense that either the LED’s or the buzzer will be able to do this? Alas I’m not smart enough to puzzle through what a solution looks like.
(original)
Replying to @tahnok
The latter of those was the subject of some failed experimentation last week. I tried emitting three tones for high, low and a return to zero, and an FFT on my computer to listen in and decode. The frequencies were visible, but I couldn’t get to “Hello,” much less “Hello, world!”
(original)
Replying to @tahnok
I’m unsure about UV; I think plain glass is enough to block it, so I assume the LCD glass would do the same? Worth trying though! re: phototransistor, I was surprised that much light made it through, but the LED backlight shines to the front, so I figured something might make it!
(original)
Replying to @josecastillo
Day 8: a trip to the auto shop; four new brake pads.
(original)
Replying to @josecastillo
I also realize that the build of Movement I flashed on this watch last month lacks all of the latest low power improvements. Still, to keep the experiment rigorous, I’m not going to take it apart to update it. This watch will run this build until it’s dead.
(original)
Replying to @josecastillo
Battery test, day 30: still at 2.97 volts! Tho I sense it’s closer to 2.965; the reading fluctuates between 2.97 and 2.96 V. Fun fact: the watch I’ve been using for accelerometer tests is down at 2.65 V and still works great! But we won’t get that far down the curve anytime soon.
(original)
Replying to @josecastillo
the dark lord Satan Claus
(original)
Replying to @josecastillo
I couldn’t help myself: I came back to ask the “festive” AI image generator for some more macabre Christmas scenes. https://app.wombo.art
(original)
Replying to @josecastillo
I know these categories are pretty basic; that’s not the point. The point is the documentation and UI took a complex process and made it simple and accessible, and I feel confident that with more time and training data I can make this tech do powerful things. That’s no easy feat.
(original)
Replying to @josecastillo
Okay this is just staggeringly cool. @EdgeImpulse has made this so easy it’s fun! From account signup to trained neural network on a microcontroller in one afternoon? Wow wow wow. This is technology at its best.
(original)
Replying to @josecastillo
Testing the neural network. When I click the button, the PyGamer strapped to my wrist streams data out to Edge Impulse, and a machine learning model classifies it. (I trained it about an hour ago, with just about five minutes of data collected the same way). This is COOL.
(original)
ok I’m gonna be honest, I am not a machine learning person. but this @EdgeImpulse tutorial is making it make a lot of sense for me. (note, this isn’t the watch; today I’m pacing the workshop with an @adafruit Pygamer strapped to my wrist) https://docs.edgeimpulse.com/docs/continuous-motion-recognition
(original)
Replying to @tahnok
If you’re interested, the watch face is actually running within Movement! The source code is here: https://github.com/joeycastillo/Sensor-Watch/blob/c5400e437f919843e0235ed66f7840fd3ddf6ea0/movement/watch_faces/demos/lis2dh_logging_face.c
(original)
Replying to @tahnok
As of now, I’m displaying it on the LCD and manually typing into a spreadsheet. It’s not the most efficient way, but the UI works: alarm button moves backwards through samples, light button cycles through which axis we’re viewing; e.g. this is 297 events on the Y axis at 1:30 PM.
(original)
Replying to @josecastillo
WOW OKAY, glad I burned 15 minutes walking to get coffee. 308 interrupts on my walk, but would you look at that: they’re almost exclusively on the Y-axis! It makes sense; arms swing at sides on a walk, but move up and down on a run. Didn’t realize it would be this pronounced tho!
(original)
Replying to @josecastillo
(the current temperature + GPIO board; Lego minifig for scale. These test points are small!)
(original)
this OTOH does feel like one of the prettier sensor boards. A second attempt at the temperature + GPIO board, it breaks out the I²C bus and TWO pins, in addition to the thermistor for temperature sensing. This could be the one for the campaign (which I’m working on today). https://twitter.com/josecastillo/status/1468059425386770434
(original)
Replying to @GeekMomProjects
that’s so very sweet :)
(original)
Replying to @GeekMomProjects
this is funny because I was *just* talking to my sister about how I’d be a terrible dog owner 😬
(original)
Replying to @josecastillo
Ugh realizing I need to go on like a 20 minute walk today, just to see the difference between a walk and a run. Honestly who has that kind of time?
(original)
Replying to @josecastillo
🎵 It’s the most wonderful end of the world… 🎵
(original)
Replying to @EdgeImpulse
heck yeah! Today the plan is to do some data forwarding with a PyGamer instead (which has an LIS3DH that’s similar to the accelerometer in the watch), but I sense it’ll be a similar process with this board, to get real data out of the watch while on a jog or a workout machine :)
(original)
Replying to @AlexGilbertson
ok that’s pretty rad!
(original)
Replying to @josecastillo
the “festive” style is the best though; it can conjure up something straight from a horror film and then, y’know drape some Christmas lights on it.
(original)
this thing is unsurprisingly fun. I had it take a crack at drawing my Oddly Specific mascot (a laser bear with a shark for an arm) and this is actually a pretty cool interpretation. https://twitter.com/adafruit/status/1468232920850649102
(original)
Findings from another morning run. For a few days I’ve been tracking my wrist’s acceleration on 3 axes. Noticing (from one experiment at least) that on a run, X events are measurably more prevalent than Z events, whereas in everyday walking around they’re more evenly distributed.
(original)
Replying to @josecastillo
Day 7: a morning run in 36° brooklyn weather. someone scared a group of birds in the street in front of me; they flocked away in perfect unison. it was really pretty.
(original)
Replying to @josecastillo
Also taking this opportunity to move from the LIS2DH to the LIS2DW. They’re both unobtanium, but I have three of the LIS2DW on hand. It’s newer; power consumption seems to be even lower than the 2DH; and the footprint seems not nearly as impossible for mere mortals to hand-place.
(original)
Not the prettiest sensor board I’ve designed, but I’m realizing if I’m gonna do this #tinyML thing, I’m going to need a way to get gobs of data out of the watch in real time while doing activities. Exposing UART test points should let me run thin data wires out of the watch case.
(original)
Replying to @jameswood
funny thing, I rewatched some of The Martian on a plane ride today, and when they blow up the airlock they actually plug the bomb into a USB-A port in the lighting panel. imagine if they forgot to pack an adapter; Matt Damon would have floated off into space forever.
(original)
Replying to @josecastillo
(it’s probably for the best; I’ve been up since 4am local time; so fried I fell asleep slumped against a pole on the G train this afternoon. I can sleep now and attack this tomorrow)
(original)
got it in my head to do some machine learning stuff before bed tonight, then realized i can’t because i don’t have the right kind of USB dongle here. this is a good future we have built for ourselves. i am fine with this.
(original)
Replying to @josecastillo
Day 6: a flight to New York.
(original)
Replying to @nebelgrau77 and @EdgeImpulse
Jackpot!
(original)
Replying to @bitshiftmask
Hey if working on this is what gets me back on my morning runs, I won’t even be mad :)
(original)
Replying to @josecastillo
I need to figure out the state of ML on Cortex M0+ class chips. I was using TensorFlow Lite for voice commands on the Open Book a while back, but that was a SAM D51; I don’t know whether the L22 in the watch is powerful enough to classify different kinds of motion from raw data…
(original)
Replying to @josecastillo
It makes sense: I’m counting interrupts where the wrist’s acceleration exceeds 1.5 g on any axis, but the smooth elliptical motion of hands & feet is designed to let you work out without any impacts that would shake up your extremeties. Still, hey, at least now I have data on it!
(original)
My folks have an elliptical machine here at home, so of course I’m trying it out with the out new Sensor Watch accelerometer code. Unfortunately, my current motion tracking parameters can’t distinguish an elliptical workout from being clinically dead.
(original)
Replying to @josecastillo
Day 4: a (vaccinated, outdoor) garden party with familia that hasn’t gotten together in years.
Day 5: an afternoon with grandma.(original)