« first day (352 days earlier)   

9:44 AM
What is (or should be) cut of a dict?
 
currently it's nothing (k7/k9). drop is defined, (vector) cut could be:
d:`a`b`c`d!1 2 3 4; 1 3_d → (`b`c!2 3;`d!4) not sure if it's useful
 
Fast Differentiable Sorting and Ranking (arxiv.org) news.ycombinator.com/item?id=22393790
 
10:14 AM
differentiable programming is a total blind spot for me reddit.com/r/apljk/comments/f2d61j/…
 
11:07 AM
@chrispsn I guess that if we stick to the lists are maps theory, cutting a dict should be no different to cutting a list, you would cut the domain and the range
 
@yiyus is the input symbols or int indices (as ktye suggested)?
 
I think that for consistency they should be integer indices, but I am not sure that would be useful for anything. k usually values more convenience than consistency
 
11:46 AM
@ngn i'm trying to compile ngn with tcc. it complains that this: ci[i][2]=al(t[i].l) is a division by zero in constant.
Any idea? maybe it's something completely different.
@yiyus we were dicussing a typo in you document: https://chat.stackexchange.com/transcript/message/53613885
is that the latest version (linked somewhere in hn)?
 
@ktye: if you want i can give you write permissions so you can fix/add anything you want
 
in oK, windows of -i gave a sliding window of size 3 with zero padding. k7 did the same with zero window. do you reckon this was deliberate?
2019-09-25 14:44:44 2core 3gb avx2 © shakti l test
 0':!4
0 0 1
0 1 2
1 2 3
2 3 0
 
12:01 PM
@yiyus thank's but i don't feel qualified. i only skimmed through it (via your doc) which looks like a major undertaking. due to kernighan, i should be twice as smart as arthur in order to understand it anyway.
@chrispsn maybe for symmetric differentiation ;)
@chrispsn it must be an important concept. some even call their company after it.
 

« first day (352 days earlier)