mlevental

11 ms·

91.
by mlevental 7y ago
elegant is a fine word: "the quality of being pleasingly ingenious and simple; neatness."
92.
by mlevental 7y ago
elegant is fine. beauty and elegance aren't the same thing. >Characterizing this appreciation as perverse and inhuman is essentially taking offense to someone else's enjoyment in something they appreciate. Seems pointless. it&#
93.
by mlevental 7y ago
this is a pet peeve of mine - the use and abuse of humanist terms in stem. you see this a lot in a math - beautiful proofs, beautiful formulas, etc. I think it's a really weird perverse even thing to look at something purely formal (me
94.
by mlevental 7y ago
https://www.youtube.com/watch?v=eJ3RzGoQC4s
95.
by mlevental 7y ago
you've been rightfully skewered by other people but I just want to point that this is what happens when someone reads one article about a controversial thing they don't otherwise know anything about and then goes around presenting
96.
by mlevental 7y ago
I love when people simultaneity have ignorant bias and reveal their biases come from complete fiction. 1 not at all people that live in poor countries (have smuggling and intermittent power) are amoral - some of them actually believe in the
97.
by mlevental 7y ago
the second paragraph spells it out? at the jump the series converges to the average rather than either value.
98.
by mlevental 7y ago
I'm interested in this. does anyone know in general where the bottle necks are on a forward pass in pytorch?
99.
by mlevental 7y ago
I don't understand? It's a piecewise differentiable function and you maximize how you maximize any such function: do gradient ascent where it's differentiable and compare against values at the boundary points (ie start, end o
100.
by mlevental 7y ago
lol this is literally by the group that's rewriting tensorflow in Swift https://www.tensorflow.org/swift so you're off on the intention here in that it is exactly taking aim at python as the main data ecosystem la
101.
by mlevental 7y ago
he/she addressed that - the points at which the function isn't differentiable has measure zero. besides this isn't some kind of new hack - one sided limits (and therefore derivatives) were invented exactly for such cases (min
102.
by mlevental 7y ago
notice the qualifier "really". obviously you can implement autodiff kind of outside the complier since pytorch and tensor flow exist. but those implementations constrain you to a select few compositions (please no comments on Turi
103.
by mlevental 7y ago
is there something in the proposal that addresses named tensors? ctrlf doesn't find anything.
104.
by mlevental 7y ago
they used to have an open platform so you could bring your own flight controller. the most recent gen of their platforms locks you into their flight controllers (and their shitty api).
105.
by mlevental 7y ago
why is this upvoted? it's completely wrong.
106.
by mlevental 7y ago
sorry i haven't seen many race cars burn out like candles or burn out like overworked employees. i have seen many though perform burnouts where they warm up the tires to increase how sticky they are.
107.
by mlevental 7y ago
i'm imagining something like the sqlite for mongo. ie you can do aggregations and nested k/v queries and filters and things like that.
108.
by mlevental 7y ago
as far as i could tell you have to jump through some hoops to use it (compile extension from source?). and i couldn't tell if it supported querying
109.
by mlevental 7y ago
asked this on a sqllite thread before but: anyone know a single file nosqldb that has good python bindings. i know I could just write to json or something but I'd like something that supports queries. right now I'm using MongoDB a
110.
by mlevental 7y ago
I don't know about the others but to burn out as the correct verb and to burnout as the incorrect verb mean two different things. I'm pretty sure this exactly the same phenomenon that gives us things like "to Google" whe
111.
by mlevental 7y ago
does anyone know a good on disk nosql db like sqlite with python bindings? I know sqllite has a json extension but I don't think you can query
112.
by mlevental 7y ago
>You are ignoring my point. No one is doubting that Minsky associated with Epstein. i am not. your point is that since he was a famous academic his name was as easily recognizable as dershowitz and pinker, who are generally famous. >W
113.
by mlevental 7y ago
well i wasn't expecting pytorch to implement (since pytorch is a graph compiler (or something like that) and therefore low-level) i was just wondering if there was something exactly like what you've linked to. so thanks!
114.
Ask HN: How to factor DL training pipelines
1 points by mlevental 7y ago|2 comments
115.
by mlevental 7y ago
lol what a rant.
116.
by mlevental 7y ago
you're going to get downvoted the all the protestant work ethic people that think labor is virtuous per se - "an honest day's work is reward in and of itself" type thing (forgetting or conveniently omitting that these va
117.
by mlevental 7y ago
You're howling at the wind. People on this forum, and in much of the rest of the world, have zero ability to empathize with someone that basically isn't their carbon copy. So you will never convince the person that you're res
118.
by mlevental 7y ago
dershowitz is a public figure. he's constantly on national talking heads news type shows and has had very public spats (thereby making him more visible) with other talking heads. Pinker, while not as famous as dershowitz, is also a pub
119.
by mlevental 7y ago
yup. as odious as palantir is blueprint is fantastic - beautiful, complete, and well documented.
120.
by mlevental 7y ago
what is "legacy ETL" exactly? you have a transpiler but what code does it transpile? what if all my ETL is written in Haskell and SQL (it's not - I don't have any ETL processes - I'm just trying to figure out what y

More ›