yvdriess

48 ms·

421.
by yvdriess 8y ago
The API does not give the bots extra information compared to a human player. The micro- and reaction time edge is also being dulled to being more human-like. They still have superhuman team fight execution though. "We’ve increased the
422.
by yvdriess 8y ago
TensorFlow is indeed the one-eyed king here. We found that running with large SparseTensors, CPU training jobs is actually faster than on GPU nodes. An individual job is faster on GPU, but SparseMatMul just blows up GPU memory use to the po
423.
by yvdriess 8y ago
I wholeheartedly agree. Partly because of that, it is too easy in SciPy to accidentally trip a conversion to a dense representation. As users typically only grab towards sparse matrices because their data otherwise wouldn't fit in memo
424.
by yvdriess 8y ago
I think it's fair to say quantum computers don't exist yet, but there is plenty of hardware. Although we're still at the wall-throwing stage and nothing seems to have stuck yet. Even for the theoretical models, we have plenty
425.
by yvdriess 8y ago
I wouldn't be surprised if their game state ran completely on integer types to avoid the awful x87 instructions.
426.
by yvdriess 8y ago
Were there collisions between players and players/NPCs? Or only on terrain.
427.
by yvdriess 8y ago
This might be a side-effect of CPUs being much less deterministic than the older hardware, leading to even harder to track down out-of-sync bugs.
428.
by yvdriess 8y ago
This. I have been looking for a company to produce me rhombus-shaped tiles to do Penrose tiling[1]. Manual labor and gearing costs just make it prohibitively expensive for smallish batches. In the end, I'm just going to cast cement ti
429.
by yvdriess 8y ago
Why not, it's on Chinese soil after all ;)
430.
by yvdriess 8y ago
Entirely speculation, but that is typically what you do when finding a cause of the RL/NN model's behavior :) If you consider the action state space, removing the 'do-nothing' state could provide a learning benefit. Cons
431.
by yvdriess 8y ago
Writing non-trivial macros in Scala or Nim will quickly liberate you of this notion.
432.
by yvdriess 8y ago
I am tired of seeing this programmer's meme constantly being repeated. Superfluous parens: Do not assume that the parentheses are superfluous, they are an excellent way to write down and communicate a tree-like structures. Code happens
433.
by yvdriess 8y ago
Any language adopting s-expr would instantly be called a Lisp. Arguably, the s-expr syntax is what enables the rich code-as-data metaprogramming that you find in Lisps. Working with full macros in a sexpr language is already complex enough,
434.
by yvdriess 8y ago
I would say 'by-demand lazy' failed. The by-demand implementation gives a guarantee that an expression is never evaluated if it is not used. Other non-strict forms exist where every expression will be touched, but no guarantees ar
435.
by yvdriess 8y ago
A supercomputer project the scale of the K-computer in Kobe.
436.
by yvdriess 8y ago
While there isn't special form for tailcall, they are not implicit. If the recursive call does something with the result, it's not a tailcall. Unless you are being cute with closures, it's already explicit in the AST. --&
437.
by yvdriess 8y ago
It's a blurry line, as with TensorFlow the heavy lifting is done in C/C++ with a python/JS wrapper. The issue becomes efficient data management in the language itself, which arguably is why numpy turned virtually into a paral
438.
by yvdriess 8y ago
That works only in limited cases, viz. where B matrix can stay in column-major format afterwards, ponder A * B * C.
439.
by yvdriess 8y ago
Dense matrix algebra has the benefit that besides the matrix dimensions, the data itself has no impact on performance: the access pattern is entirely predictable and known as soon as you know the matrix sizes (side note: which is why just-i
440.
by yvdriess 8y ago
Plenty of overlap. Ipad stream-gaming on the couch while the wife watches TV, for example.
441.
by yvdriess 8y ago
It beats it in the same way that procedures beat subroutines that use a stack.
442.
by yvdriess 9y ago
It is useful in case where the hardware prefetcher is not helpful. That is, indirect memory access (e.g. graph traversal), constantly accessing different pages (eg column wise access of a matrix with >4k kB rows) Howeve, finding the rig
443.
by yvdriess 9y ago
Zach, if you're reading this, please make a Brainfuck expansion for TIS-100 [1] and Shenzhen I/O [2], your games are clearly not hard enough already ;) [1] http://www.zachtronics.com/tis-100/ [2] http:/
444.
by yvdriess 9y ago
Writing clean C loops are a good way of writing vectorized code. Checking the vectorization report of icc 2018 for fprop() shows that the loops got inlined, unrolled and vectorized (with AVX512 and AVX2 instructions using XMM/YMM regis
445.
by yvdriess 9y ago
It's commonly optimized out, yes. Rule of thumb is typically to pass pointers to a struct if it gets fairly large and complex. Even then, you should really profile to see that the becomes a bottleneck, compilers inline aggressively and
446.
by yvdriess 9y ago
There is a qualitative difference between cellphone use during school hours and sitting behind a laptop on your own time. Most HN visits are probably more of the latter.
447.
by yvdriess 9y ago
Lucid is a functional programming language from the '80s that took a very different approach. All variables are infinite streams of values. It is a grandfather of sorts of reactive programming. Although the book title claims it's
448.
by yvdriess 9y ago
People often say that one should learn Lisp/Prolog/Haskell to become a better programmer. Lucid and Multidimensional Lucid should definitely be added to that list. It just reconfigure your brain around loops and iteration spaces.
449.
by yvdriess 9y ago
Gambling is not banned in Belgium, it is regulated. There is a state-owned lottery, plenty of casino's, sportsbetting companies etc. One thing that is banned is selling gambling products to minors. You are right that booster-pack based
450.
by yvdriess 9y ago
Spoilers! ;)

More ›