acmj

9 ms·

61.
by acmj 6y ago
> at worse it may be a scam making money from donations Before making such a rude conjecture, at least check the dev's github page [1] first. Most of his git commits are related to V and he has been actively working on it for a co
62.
by acmj 6y ago
Several recent assemblers are faster than WENGAN. For nanopore, there are shasta and wtdbg2 (both published). For HiFi, there are Peregrine and hifiasm (both unpublished but with preprints). I also found their phrasing here misleading: &quo
63.
by acmj 6y ago
No, not even a single one of them. Read Gene Myers paper in 1995 or 2005. Modern OLC assemblers all follow that route which has nothing to do with the Hamilton problem. Equating overlap based assembly to a Hamilton problem is the biggest li
64.
by acmj 6y ago
No practical assemblers take assembly as a Hamiltonian problem.
65.
by acmj 6y ago
HiCanu achieved that a year ago. WENGAN is a worse assembler with clearly more misassemblies. It amazes me that this level of paper can be published in Nature Biotech.
66.
by acmj 10y ago
I later realized he is the developer, but this does not change this discussion. Here is a micro benchmark, computing softmax 1 million times over a random vector of size 1000. On an old Linux server, calling the libm expf once takes 11.76 C
67.
by acmj 10y ago
Thanks for the pointer. The full softmax implementation is here [1]. I have not read the code, but I can trust the developer to have a very fast implementation. Nonetheless, I don't think the reference implementation in your original l
68.
by acmj 10y ago
Where is this vectorized expf implemented? I am only seeing softmax is calling the standard expf. Let's suppose expf from libm is vectorized. Is there any benchmark showing nnpack's implementation is really faster? I doubt, actual
69.
by acmj 10y ago
In the nnpack implementation, the same exponential (i.e. expf) is computed twice for each element, which is a waste of time. A faster implementation should save each expf result to output[sample][channel] first, compute the sum and then res
70.
by acmj 10y ago
In the past several years, quite a few developers have tried GPUs for analyzing bio-sequences, but found the speedup is modest. Good GPUs are expensive. It is usually better to put that money on CPUs or RAM.
71.
by acmj 10y ago
Yeah, I heard about it from others, too. Look forward to it. Hope the dev team stick to the plan.
72.
by acmj 10y ago
Julia is great, but IMO, it needs to reach 1.0 first before making a splash. It would be frustrating if you found your tools developed on v0.4 stopped working on v0.5 released a year later. API/ABI stability is critical to the adoption