cshenton

13 ms·

91.
by cshenton 9y ago
It's probably not correct to say the stance is insincere. If it is purely profit motivated, then it would be quite sincere of Apple to say they care about privacy, in the same way it's sincere of them to say they care about qualit
92.
by cshenton 9y ago
I'm somewhat doubtful of the design as it's explained in this blogpost, seems like there will be lots of treatment externalities. I'd really recommend anyone who's interested read http://cega.berkeley.edu&#x2F
93.
by cshenton 9y ago
So, no change then? Since those "home wifi" connections come through a centrally owned comms network.
94.
by cshenton 9y ago
That's not true. A requirements file is the canonical way to do this. Just use a requirements file and pip install -r requirements.txt. Then version control that file and other Devs do the same. Just like a gemfile.
95.
by cshenton 9y ago
If you get into the habit of making a virtual environment and requirements.txt in every repo then it's smooth. Just activate the env when you're using the repo and use vanilla pip install to install dependencies. Then pip freeze t
96.
by cshenton 9y ago
When it comes to a name for this whole situation, I always liked the image of 'cryptographic tulips'.
97.
by cshenton 9y ago
Increasing in recent years. For a long time, one had to use mathematical shortcuts to make the techniques computationally feasible (for example conjugate priors). Now, with better sampling algorithms (like NUTS) you don't have to make
98.
by cshenton 9y ago
Thats not an option when you want to do something like reinforcement learning with lock free updates. In that case, the networks are small enough that you want to use the CPU, but learning sensitive enough that you don't want multiple
99.
by cshenton 9y ago
So did I. And every time I downloaded it again, I'd open a medium sized file, scroll down, the scroll lags, and I delete it again.
100.
by cshenton 9y ago
50% seems like pretty optimistic odds for an angel investment.
101.
by cshenton 9y ago
In what circumstances is it only 10x slower? Random search is totally useless when your environment is stochastic. These algorithms aren't learning sequences of actions, in fact most use a 30 'no op' random start to avoid jus
102.
by cshenton 9y ago
There's also the fact that, for high level languages like python, the gRPC server has lower throughput than available python http servers. Even with the overhead of parsing http and whatever message format you're sending over it.
103.
by cshenton 9y ago
Because bson, msgpack etc are self describing? It means a way lower barrier to entry.
104.
by cshenton 9y ago
Do you think most devs spend enough time thinking through a good way to model what they're building?
105.
by cshenton 9y ago
This is super cool. Might be replacing something I've been building with it. Are there any benchmarks I could check out? Especially for a3c.
106.
by cshenton 9y ago
So it can train on one game and play without training on a previously unseen (but also atari) game? That's pretty neat, DQN and A3C certainly can't do that.
107.
by cshenton 9y ago
This is super cool, but it doesn't outperform deep learning based RL methods. In fact, I'm not sure how much more compute efficient than something like A3C it would be. That can produce 4x the score of DQN in a comparable number o
108.
by cshenton 9y ago
I've found falcon a real pleasure to work with building pure json APIs, super sane approach to hooks and middleware which makes handling auth and db pooling both explicit and simple. Plus it's pretty fast which is always nice for
109.
by cshenton 9y ago
My apprehension about using flask for microservices is that calls to another microservices are going to block. With NodeJS, the webserver and http libraries are async and mature. Python's really missing a solid async requests library (
110.
by cshenton 9y ago
Be careful with aiohttp, async io is great, but the actual http parser in aiohttp is super slow. See https://magic.io/blog/uvloop-blazing-fast-python-networking/ for a full write up. I've been building stuff
111.
by cshenton 9y ago
Except that under most contracts, the company already owns the code currently used to automate the job.
112.
by cshenton 9y ago
Didn't tflearn become tensorflow.contrib.learn? So there's technically a high level API in the TF standard library. Though Keras would probably be a better choice if OP wanted a high level API for deep learning.
113.
by cshenton 9y ago
As someone who ended up majoring in pure mathematics, I always thought it was a shame that theres this wall of arithmetic and calculus between you and the first pure mathematics class. I suck at those, but I do well with abstractions and pr
114.
by cshenton 9y ago
Federated Learning. It's something Google does at the Android level, but I don't think they provide an API for app developers to train on device. Another use case is privacy: ship a pretrained facenet style model and let it learn
115.
by cshenton 9y ago
You may be thinking of anthropocentrism.
116.
by cshenton 9y ago
Probably because it's pretty bleeding edge. Google Cloud Endpoints, which released earlier in the year, allows you to write a gRPC server, and offers the HTTP proxy as part of the service. The ecosystem still isn't quite there tho
117.
by cshenton 9y ago
What problems do you run into? I've had good luck with a pretty vanilla virtualenv + requirements.txt files, and docker for packaging (great for c libs that my python packages call down to).
118.
by cshenton 9y ago
gRPC generates rpc server and client stubs based off a protocol buffer definition. Saying it's a replacement for REST makes little sense since it's possible to define a REST API within it. It's also possible to write a totall
119.
by cshenton 9y ago
Totally disagree from the user's perspective. Every time I try and use Quora for technical reasons, the answers are almost always people selling their product and not offering a constructive solution to the problem asked. Sure, if the
120.
by cshenton 10y ago
Yeah Google has done this in the past. It's what they did with Google Wave. But Lazlo Bock has reflected that this caused the a lot of departures when the project failed. Not sure if they do that anymore.

More ›