timanglade

10 ms·

31.
by timanglade 7y ago
Archipelago | Software Engineers, Product Managers | San Francisco or REMOTE (US only) | Fulltime We're an early stage startup, still in stealth, working to change how risk is insured. Our founders are tech & finance entrepreneurs
32.
by timanglade 7y ago
Archipelago | Software Engineers, TPMs | San Francisco or REMOTE (US only) | Fulltime We're an early stage startup, still in stealth, working to change how risk is insured. Our founders are tech & finance entrepreneurs with several
33.
Diving Deeper into Consensus(blog.helium.com)
60 points by timanglade 8y ago|4 comments
34.
From Zero to Azure IoT in Five Minutes(blog.helium.com)
5 points by timanglade 9y ago|0 comments
35.
by timanglade 9y ago
Just voicing your interest here is fine!
36.
by timanglade 9y ago
Just US & Canada, for now.
37.
by timanglade 9y ago
Ha seems like a fun coincidence. The writers came up with it early in the writing of season 4, and I started working on it sometime in the Summer of 2016 iirc. As far as the origin story goes, it was just great writers coming up with a grea
38.
by timanglade 9y ago
Oh neither, there is nothing to find in the binaries :)
39.
by timanglade 9y ago
It’s fair, as I mention in the blogpost there are some failures that are a bit obvious for sure. Mostly I think I tried to fit too many things into one “hotdog” category including chili dogs, chicago dogs, bunless hotdogs, cut up hotdogs, e
40.
by timanglade 9y ago
Not my work by any means, this is all community-driven, and I think they do as awesome a job as is possible to do, considering the constraints Apple puts in their way. This guide has all the steps: https://egpu.io/setup-guid
41.
by timanglade 9y ago
The amazing react-native-camera plugin! [0] I’m still getting a few camera-related crashes on Android right now, but overall I would say it makes things pretty smooth! [0]: https://github.com/lwansbrough/react-native-ca
42.
by timanglade 9y ago
Biggest regret was not keeping a pristine dataset for final testing / evaluation on device. I ended up flying blind when it came to setting the final threshold, testing the effects of quantization, or even just measuring the distortion
43.
by timanglade 9y ago
I can’t recommend Rachel Thomas and Jeremy Howard’s FastAI course enough! I attended it in person in SF, but the YouTube recordings and online community around it are great! [0] Beyond that, I would recommend making sure you have a concrete
44.
by timanglade 9y ago
Thanks! I really wanted to demystify as many of the steps as possible. It’s easy to see the finished result, but really it took a lot of trial & error to get something even as simple as this out there, and I wanted to make sure others f
45.
by timanglade 9y ago
Thanks! I definitely think executing neural networks on-device is the future for a lot of applications. It’s just a better UX, and much cheaper to boot!
46.
by timanglade 9y ago
Yeah I was surprised it became so fast once I started using small networks. I actually toyed with the idea of slowing down the transition to results artificially to provide better UX lol
47.
by timanglade 9y ago
I thought there might be, but most of the AI code ended up being native. The only AI code in React Native is a single line: var percentage = await NativeModules.AIManager.analyzeImage(path) … Everything below that is Java or Object
48.
by timanglade 9y ago
Great question — I did not, because I had unfortunately spent all of my data on that last training run, and I did not have a untainted dataset left to measure the impact of quantization on. (Just poor planning on my part really.) It’s also
49.
by timanglade 9y ago
While we’re here and chatting about this, I should say most of the credit for this app should really go towards the following people: Mike Judge, Alec Berg, Clay Tarver, and all the awesome writers that actually came up with the concept: Me
50.
by timanglade 9y ago
I’m not sure, I think I would maybe break classes into multiple labels, but that becomes even more finicky to train. At the end of the day, there are many more things that are not hotdogs, than things that are hotdogs, so you do have to pro
51.
by timanglade 9y ago
My takeaway is that local development has a huge developer experience advantage when you are going through your initial network design / data wrangling phase. You can iterate quickly on labeling images, develop using all your favorite
52.
by timanglade 9y ago
I’m glad to hear the app is an inspiration for real — and more meaningful — apps ;D I was originally inspired to take this on by Pete Warden and his TensorFlow for Mobile Poets approach, and nothing would make me happier than to see this ap
53.
by timanglade 9y ago
Yes, that’s what you see in the picture, although as completely personal advice, I would stop short of recommending it. For one there are arguably better cases out there now, and you can sometimes build your own eGPU rig for less. Finally,
54.
by timanglade 9y ago
Well for a while I was lulled into complacency because the retrained networks would indicate 98%+ accuracy, but really that was just an artifact of my 49:1 nothotdog:hotdog image imbalance. When I started weighing proportionately, a lot of
55.
by timanglade 9y ago
Lots of manual searching, vetting & labeling! Definitely the most actively time-consuming part. (Passively, only the wait between training runs was longer.)
56.
by timanglade 9y ago
Not offtopic at all! Dirty hack for sure. The enclosure I bought was a hack, the drivers were a hack, and there was software on top that was a hack as well. But the developer experience was totally awesome… Almost made the constant graphics
57.
by timanglade 9y ago
Yup and in fairness maybe that’s something the community (myself included) should really step in and improve — but it’s not always clear how the leadership of the project would like these things to improve, and I often get the feeling they
58.
by timanglade 9y ago
Ha, I’d love to hear more. What was the app for? I can’t imagine why you’d have to pick up on Trebek’s elocution??
59.
by timanglade 9y ago
a gentleman never tells
60.
by timanglade 9y ago
SGD with Cyclical Learning Rates [0]. Honestly, it’s the closest to a Machine Learning silver bullet I’ve found to date! That paper is awesome . [0]: https://arxiv.org/abs/1506.01186

More ›