These are chat archives for symengine/symengine

14th
Mar 2017
Isuru Fernando
@isuruf
Mar 14 2017 14:54
@srajangarg, use ostringstream instead of std::to_string
berceanu
@berceanu
Mar 14 2017 21:31
@isuruf
do you know a good book on modern scientific C++?
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 21:52
Wasn't a new standard just passed?
Francesco Biscani
@bluescarni
Mar 14 2017 21:53
there's barely books about general-purpose modern C++, I doubt there's much about modern scientific C++
@ChrisRackauckas C++17 has been finalised, but technically it's not a standard yet
berceanu
@berceanu
Mar 14 2017 21:55
C++boost++
@ChrisRackauckas been digging a bit more into that CERN code. i found 3 different ad-hoc runge kutta implementations
in the same condebase!
@bluescarni website at least? or examples of good code?
Francesco Biscani
@bluescarni
Mar 14 2017 22:01
is depends mainly what you mean by "modern C++" and "scientific" :) I mean, if you are interested in linear algebra, numerical integrators, etc. there's not so much in modern C++ of interest. Modern C++ shines especially in generic/template (meta)programming, but at the end of the day most scientific code does not care about that. Most people seem interested only in double and arrays at the end of the day
berceanu
@berceanu
Mar 14 2017 22:02
but then they might as well use C
Francesco Biscani
@bluescarni
Mar 14 2017 22:02
they do still use fortran :)
berceanu
@berceanu
Mar 14 2017 22:03
fortran is a DSL created for numeric computing
it actually makes sense to use it, at least until it can be completely replaced by Julia
Francesco Biscani
@bluescarni
Mar 14 2017 22:03
Boost stuff is high quality, but not much of it could be described as modern C++
for a taste of modern stuff you can look at something like Boost.Hana, which is strictly >= C++14
berceanu
@berceanu
Mar 14 2017 22:05
"For the above code to actually work, std::array's operator== would have to be marked constexpr, which is not the case (even in C++14)."
Francesco Biscani
@bluescarni
Mar 14 2017 22:05
my projects are all in >= C++11, but probably not the flavour of scientific computing people mean when talking about scientific computing
berceanu
@berceanu
Mar 14 2017 22:05
well, let's wait for C++25
well, give me an example and i can tell you if it's what i mean by scientific computing :)
Scott P. Jones
@ScottPJones
Mar 14 2017 22:07
I'm just thankful that Julia keeps me from from having to go back to all the complexity of C++ (even 14 or 17, which yes, are better than older standards that I had used)
Francesco Biscani
@bluescarni
Mar 14 2017 22:08
this is a small project I decoupled from my main one https://github.com/bluescarni/mppp it's a small wrapper around GMP with faster arithmetics for small values (1 and 2 limbs)
I use it for my CAS library for Celestial Mechanics
it's not heavy on modern C++ stuff (it has to compile on MSVC 2015 unfortunately), but it does use modern idioms
my main project is this one:
https://github.com/bluescarni/piranha
it is woefully undocumented though
(in terms of user docs, the API is 100% documented)
and the last month I have been full time on this:
which is the re-incarnation of an older code I wrote now a few years ago
berceanu
@berceanu
Mar 14 2017 22:14
i see, thank you!
Francesco Biscani
@bluescarni
Mar 14 2017 22:15
sure, no problem
berceanu
@berceanu
Mar 14 2017 22:15
it's not exactly what i had in mind, but i think it can be a good resourse for learning
btw, i see you worked at Schlumberger and ESA previously
Francesco Biscani
@bluescarni
Mar 14 2017 22:15
yep
berceanu
@berceanu
Mar 14 2017 22:15
how was it? :)
Francesco Biscani
@bluescarni
Mar 14 2017 22:16
ESA or SLB?
berceanu
@berceanu
Mar 14 2017 22:16
i mean, compared to academia
both
Francesco Biscani
@bluescarni
Mar 14 2017 22:16
the position I had in ESA was full research, in this place: http://www.esa.int/gsp/ACT/index.html
was my first job after the phd, fantastic experience
SLB of course was different :) pretty stereotypical corporate experience... or at least it lined up to what I expected from the environment
I stayed for little more than one year at SLB, then moved country and went back to an academic environment
berceanu
@berceanu
Mar 14 2017 22:19
so the ESA thing was also a postdoc?
Francesco Biscani
@bluescarni
Mar 14 2017 22:19
yes, 3 years
berceanu
@berceanu
Mar 14 2017 22:19
i see
i had the preconceived idea that it's easier to get a permanent postition there than in academia
Francesco Biscani
@bluescarni
Mar 14 2017 22:20
me too, before getting there. the reality of things at ESA was quite different, it's mostly a giant bureaucratic apparatus
berceanu
@berceanu
Mar 14 2017 22:20
:)
Francesco Biscani
@bluescarni
Mar 14 2017 22:20
they do nearly zero research, with some rare exceptions like the team I was in
it's an agency which is basically tasked with distributing money to the "european space market"
berceanu
@berceanu
Mar 14 2017 22:21
which space market? :)
do we have one?
Francesco Biscani
@bluescarni
Mar 14 2017 22:22
:) there's a bunch of companies, a few giants eating up almost everything and a few smaller players
it's largely a subsidized market, even more when I was there... no spaceX to shake things up back then
berceanu
@berceanu
Mar 14 2017 22:22
and in SLB i guess you were also in a research team
do they do real research?
Francesco Biscani
@bluescarni
Mar 14 2017 22:23
I was working as HPC software engineer on a oil reservoir simulator
berceanu
@berceanu
Mar 14 2017 22:23
did they have millions of lines of legacy code?
Francesco Biscani
@bluescarni
Mar 14 2017 22:24
exactly, and this was the next generation C++ code which was already rotting before becoming profitable
berceanu
@berceanu
Mar 14 2017 22:24
i see
Francesco Biscani
@bluescarni
Mar 14 2017 22:24
I was in a small group that investigated the use of GPU computing and similar technologies to accelerate the simulations
berceanu
@berceanu
Mar 14 2017 22:24
well, sounds fun
CUDA stuff?
Francesco Biscani
@bluescarni
Mar 14 2017 22:25
yes.. but in reality we were really integrating code rather than writing much new stuff
berceanu
@berceanu
Mar 14 2017 22:25
didn't get a chance to play with that yet
Francesco Biscani
@bluescarni
Mar 14 2017 22:26
they had decided to outsource the hard CUDA bits to a separate company in maryland
berceanu
@berceanu
Mar 14 2017 22:26
it's funny that you say "typical corporate environment"
Francesco Biscani
@bluescarni
Mar 14 2017 22:26
Scott P. Jones
@ScottPJones
Mar 14 2017 22:26
Did you ever deal with a project at the ESA to collect and store data from many of their satellites (in a database)?
berceanu
@berceanu
Mar 14 2017 22:26
because i have no idea what that's like
Francesco Biscani
@bluescarni
Mar 14 2017 22:27
@ScottPJones no our team's task was to think about the 20/30 years space research horizon rather than working on specific missions
berceanu
@berceanu
Mar 14 2017 22:27
wow, there's companies specialized in reservoir simulation
Francesco Biscani
@bluescarni
Mar 14 2017 22:27
there's huge money in that, everything that has to do with oil is extremely rich as you can imagine
Scott P. Jones
@ScottPJones
Mar 14 2017 22:28
Ah, OK. I worked a bit with the ESA (when I worked for the company that won the contract to store the huge amount of data coming from their satellites)
Francesco Biscani
@bluescarni
Mar 14 2017 22:28
@ScottPJones the team I worked in was composed of young researchers, often at their first postdoc
coming from all sorts of different backgrounds
it's a bit of a white fly in ESA
Scott P. Jones
@ScottPJones
Mar 14 2017 22:29
I was just amazed by the huge amount of data they were constantly receiving from all sorts of satellites
berceanu
@berceanu
Mar 14 2017 22:29
ok so my mental image of "corporate environment": deadline-driven profit-oriented development, under the supervision of (usually technically incompetent) managers; lots of beaurocracy and favoritism; little original research. am i far off?
Scott P. Jones
@ScottPJones
Mar 14 2017 22:29
(it wasn't an easy task to meet their requirements - a lot of the "big name" databases couldn't handle it ;-) )
berceanu
@berceanu
Mar 14 2017 22:30
define huge
Francesco Biscani
@bluescarni
Mar 14 2017 22:30
no you are not far off... one of my goals getting the job at SLB was to experience for myself and to test out my preconceived notions about closed source/corporate soft development and environment
let's say most of my prejudices were not far off from the reality
berceanu
@berceanu
Mar 14 2017 22:31
:)
on the flipside, it's probably a comfy life compared to academia
no need to publish or perish or change country every few years
and when you're out the door at 5, it's out of your mind as well
right?
Francesco Biscani
@bluescarni
Mar 14 2017 22:32
it's certainly great if you are looking for a relatively safe, well-paying job with very good benefits, especially in the oil sector
Scott P. Jones
@ScottPJones
Mar 14 2017 22:33
Here is a link to one of the articles about it: http://www.odbms.org/wp-content/uploads/2013/11/Charting_the_Galaxy.pdf
It was from the Gaia satellites
Francesco Biscani
@bluescarni
Mar 14 2017 22:33
@ScottPJones when I was at ESA they were looking into cloud solutions to their data problems, but I am not sure if anything came out of that
Scott P. Jones
@ScottPJones
Mar 14 2017 22:33
What years were you there?
Francesco Biscani
@bluescarni
Mar 14 2017 22:34
2009-2012
berceanu
@berceanu
Mar 14 2017 22:34
maybe it was before the "big data" revolutioin
Francesco Biscani
@bluescarni
Mar 14 2017 22:34
it was during the early cloud computing hype
berceanu
@berceanu
Mar 14 2017 22:34
hadoop, spark and all
@ScottPJones they use Java?!
Francesco Biscani
@bluescarni
Mar 14 2017 22:35
they had issues with the usual cloud providers because they mostly were not on european soil, and they were afraid about data security
so they were looking to deploy in-house cloud solutions
berceanu
@berceanu
Mar 14 2017 22:35
turns out, they were right ;)
Francesco Biscani
@bluescarni
Mar 14 2017 22:35
but I don't know how it ended
Scott P. Jones
@ScottPJones
Mar 14 2017 22:36
Their applications weren't in Caché ObjectScript - IIRC the data came in as 600 byte Java objects.
I came up with some ideas to store it in a lot less space ;-)
berceanu
@berceanu
Mar 14 2017 22:37
all you hear on the news is "renewable energy" but hey, it turns out what really pays off is simulating oil reservoirs!
Scott P. Jones
@ScottPJones
Mar 14 2017 22:38
I'm just happy I got my solar panels in before 45 took over ;-)
berceanu
@berceanu
Mar 14 2017 22:38
45?
Scott P. Jones
@ScottPJones
Mar 14 2017 22:38
Our current "alt-pres" ;-)
berceanu
@berceanu
Mar 14 2017 22:39
the crazy thing is though, with diminishing oil resources, this things are only going to get more profitable
because it doesn't look like we will switch to green stuff anytime soon
Scott P. Jones
@ScottPJones
Mar 14 2017 22:39
It's a meme here in the US to not use his name - just his number.
Francesco Biscani
@bluescarni
Mar 14 2017 22:39
president peepee? :)
berceanu
@berceanu
Mar 14 2017 22:39
he's no 45?
:D
Scott P. Jones
@ScottPJones
Mar 14 2017 22:40
Haha! Yep.
berceanu
@berceanu
Mar 14 2017 22:40
didn't realise
Scott P. Jones
@ScottPJones
Mar 14 2017 22:40
I'm waiting for those videos to "leak"
berceanu
@berceanu
Mar 14 2017 22:40
make solar great again
forget scientific software guys, let's write a reservoir simulator!
btw, i lived in the netherlands for a couple of years
Scott P. Jones
@ScottPJones
Mar 14 2017 22:41
I haven't paid for electricity for a full year now, and actually am net negative more than 1 megawatt
berceanu
@berceanu
Mar 14 2017 22:42
and there was a huge campain for renewables from Shell
Francesco Biscani
@bluescarni
Mar 14 2017 22:42
NL is a great place to live
berceanu
@berceanu
Mar 14 2017 22:42
i wonder if they're actually doing things in that direction or it's all PR
Scott P. Jones
@ScottPJones
Mar 14 2017 22:42
used to use more than 2x the electricity of my neighbors (all my computers, hot tub that we use year-round, and pool pump in the summer)
berceanu
@berceanu
Mar 14 2017 22:42
in theory, the big oil companies are also the ones that will lead in the transition to green energy
@ScottPJones you have to factor in the initial investment as well though, and all the service costs
Scott P. Jones
@ScottPJones
Mar 14 2017 22:43
yeah, right!
For me, it will break even in 5 years
because of a program where credits for the generation is sold at an auction (but with a minimum value set by the state government), for 10 years.
so I get paid almost $4000 per year for generating around 16 MWH
berceanu
@berceanu
Mar 14 2017 22:45
sweet :)
Scott P. Jones
@ScottPJones
Mar 14 2017 22:45
and I used to pay around $4000 per year.
Also, federal and state tax rebates, cut cost from $60K to $45K
So, a bit over 5 years, it will be paid off, and then I'll be saving $4K and getting paid $4K for 4.5 more years, then another 10 years of saving $4K per year (and that's if electricity prices haven't risen)
Francesco Biscani
@bluescarni
Mar 14 2017 22:48
I believe they have similar incentive programs here in Germany
Scott P. Jones
@ScottPJones
Mar 14 2017 22:48
But what's crazy, is people like the Koch brothers (and other oil interests) have blocked that sort of programs, and rebates for electric/hybrid cars, in states like Florida, that get much more sun than Massachussetts.
berceanu
@berceanu
Mar 14 2017 22:49
Koch brothers are still alive?!
Scott P. Jones
@ScottPJones
Mar 14 2017 22:49
In Europe though, from what I've heard, the electric companies have prevented the use of "net" metering for electricity.
You have to use a much less efficient battery system for when you generate more than you use.
exciting stuff
Scott P. Jones
@ScottPJones
Mar 14 2017 22:51
;-) I was born in Tulsa, Oklahoma, which used to be the Oil Capitol of the world.
berceanu
@berceanu
Mar 14 2017 22:52
:D
Francesco Biscani
@bluescarni
Mar 14 2017 22:52
@ScottPJones what do you mean exactly? that you must store the excess energy rather than push it back into the grid?
berceanu
@berceanu
Mar 14 2017 22:53
what i find fascinating about all this, is i imagine the companies doing these simulations
Scott P. Jones
@ScottPJones
Mar 14 2017 22:53
(more oil companies were based in Tulsa than anywhere else - but later on, lots of consolidation, when companies merged, they didn't keep the merged headquarters in Oklahoma, a lot ended up in Houston.
berceanu
@berceanu
Mar 14 2017 22:53
"let's buy a new cluster"
"ok, done"
whereas for us in academia it's like "ok maybe we can get a new node. let's start writing a grant application"
Scott P. Jones
@ScottPJones
Mar 14 2017 22:54
Yes, the excess energy has to be stored, not pushed back on the grid.
berceanu
@berceanu
Mar 14 2017 22:54
"ok now let's pray"
Scott P. Jones
@ScottPJones
Mar 14 2017 22:54
Tesla is really big on that tech.
Francesco Biscani
@bluescarni
Mar 14 2017 22:54
yeah I was reading about their home batteries
berceanu
@berceanu
Mar 14 2017 22:54
well they're the king of GPU
and they're really pushing it into HPC land
sorry i meant Nvidia
not tesla
Scott P. Jones
@ScottPJones
Mar 14 2017 22:55
I've also heard that one idea is to use your electric car's batteries as additional storage for energy generated during the day, letting you power your house from your car at night ;-)
berceanu
@berceanu
Mar 14 2017 22:55
tesla has the battery pack, yes
power wall, was it
Francesco Biscani
@bluescarni
Mar 14 2017 22:55
:)
they also have to roof tiles right?
berceanu
@berceanu
Mar 14 2017 22:56
the great power wall with mexico :)
Scott P. Jones
@ScottPJones
Mar 14 2017 22:56
Well, he also had the biggest solar company in the US, SolarCity, ended up merging it into Tesla
berceanu
@berceanu
Mar 14 2017 22:56
are those out already?
Scott P. Jones
@ScottPJones
Mar 14 2017 22:57
I haven't seen them available yet - I got some pretty nice panels when I had mine installed last March
berceanu
@berceanu
Mar 14 2017 22:58
C++ + CUDA + MPI looks like a mess
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 22:58
the great power wall with mexico :)
Francesco Biscani
@bluescarni
Mar 14 2017 22:58
it's a big bowl of pain yes
berceanu
@berceanu
Mar 14 2017 22:58
but i guess it's the only real option when doing simulations on a GPU cluster
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 22:59
"We're going to build a wall, and make green energy companies pay for it! And it will actually be solar panels."
berceanu
@berceanu
Mar 14 2017 22:59
:D
Francesco Biscani
@bluescarni
Mar 14 2017 22:59
GPUs are actually not that great for oil reservoir simulators
Scott P. Jones
@ScottPJones
Mar 14 2017 22:59
Why not?
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:00
Xeon Phi is where it's actually at for a lot of things.
Francesco Biscani
@bluescarni
Mar 14 2017 23:00
well first they have a memory problem, the simulations are huge (1E8 cells) and the memory on the cards is limited with respect to the memory you have on a medium server
berceanu
@berceanu
Mar 14 2017 23:00
@bluescarni did you tell that to SLB?
Francesco Biscani
@bluescarni
Mar 14 2017 23:01
I learned it there :)
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:01
oh, they are large finite difference simulations?
berceanu
@berceanu
Mar 14 2017 23:01
"they wouldn't listen"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:01
Yeah... it can be tough to use GPUs correctly for that
Francesco Biscani
@bluescarni
Mar 14 2017 23:01
they are large finite volumes/finite elements simulators yeah
Scott P. Jones
@ScottPJones
Mar 14 2017 23:01
What about Nvidia's latest GPUs? Any better?
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:01
Xeon Phis don't have as much data transfer issues as GPUs
GPUs are only good if the problem is reduced to pure linear algebra
that's pretty much it
Francesco Biscani
@bluescarni
Mar 14 2017 23:02
and secondly, the typical computational bottleneck is the solution of very large and very sparse linear system
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:02
linear algebra that fits on one GPU, that is
Francesco Biscani
@bluescarni
Mar 14 2017 23:02
@ScottPJones they have more RAM but it's till not much compared to what you have on a regular server
sparse linear systems need sparse linear algebra, which is much less of a slam dunk on the GPU than dense linear algebra
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:03
it really boils down to solving a sparse Ax=b
Francesco Biscani
@bluescarni
Mar 14 2017 23:03
yep
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:03
which builds down to multiplying vectors times A
if A fits on a GPU, you're good
berceanu
@berceanu
Mar 14 2017 23:03
hedge funds are using FPGAs
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:03
otherwise you have to A*v by splitting A across systems
but then, you have different parts of the vector possibly changed by different parts of the A
so you can't just do it all independently
so there is a sort of "map to do A*v for the part of A you have", then a "reduce to get the real full v"
at that point, data transfer speeds really matter
especially when A is adaptive.
berceanu
@berceanu
Mar 14 2017 23:04
i can't help wondering if the hardware manufacturers have their interests lined up with the scientific computing community
Francesco Biscani
@bluescarni
Mar 14 2017 23:05
@ChrisRackauckas yes, and you have the additional aspect that typical sparse algorithms for linear algebra are much more branchy than in the dense case, and GPU don't do as well with branch rich code
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:05
well, the GPU doesn't need to deal with the branches usually
it's usually just used for matrix-vector multiplications
berceanu
@berceanu
Mar 14 2017 23:05
so dense linear algebra with reasonable sized matrices, and GPUs are golden
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:05
all sparse iterative methods are just different ways to get a good approximation matrix
exactly
which is why machine learning loves them
Scott P. Jones
@ScottPJones
Mar 14 2017 23:06
I think they've been learning - it's kind of, they built these things for graphics, for games, and now everybody is doing so much other stuff with them, and they are having to totally rethink things
berceanu
@berceanu
Mar 14 2017 23:06
why are they promoted as the wholy grail then
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:06
because data science...
berceanu
@berceanu
Mar 14 2017 23:06
oh, that
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:06
yeah...
berceanu
@berceanu
Mar 14 2017 23:06
ok but the rest of science has different needs
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:06
when you say "I do reserach, and have used a computer before"
people go "data science?"
data science has more money
berceanu
@berceanu
Mar 14 2017 23:07
"i do physics"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:07
it has the funding of all of the big tech companies
berceanu
@berceanu
Mar 14 2017 23:07
"wait, is that number science?"
Scott P. Jones
@ScottPJones
Mar 14 2017 23:07
hehe ;-) I go to the Open Data Science Conferences, and there are a lot of really smart people working on all sorts of stuff that they consider "data science"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:08
I'm not saying it's bad
but it is a huge fad with lots of money that is pushing aside other parts of computational science because it is a much larger crowd
berceanu
@berceanu
Mar 14 2017 23:08
well it sort of bugs me when really smart people work on optimizing the ads i see online
it's grossly overhyped yes.
"computers are taking over"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:09
that's why you see some big HPCs (Stampede2) going to Xeon Phi clusters (kind of like GPUs, but faster data transfer so better for solving diffeqs and pdes)
and they get criticized: "but those aren't as fast for ML/neural nets"
Scott P. Jones
@ScottPJones
Mar 14 2017 23:09
what is of concern (and is much discussed at the ODSCs) is all the people who've heard it's very lucrative, and calling themselves "Data Scientists", without a lot of the background they should really have (in math, statistics, domain knowledge)
a lot of boot camps
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:09
ye
berceanu
@berceanu
Mar 14 2017 23:09
well sure
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:10
the bigger issue is that there's so much more money and jobs there...
berceanu
@berceanu
Mar 14 2017 23:10
what do you mean, you can't learn it all in 4 weeks?
import datascience
problem.solve()
done
Scott P. Jones
@ScottPJones
Mar 14 2017 23:10
maybe the ODSC conferences are a bit different, because they push for open data science, there's a lot of good stuff about how to use ds for good, avoiding the problems of bias-in, bias-out, etc.
berceanu
@berceanu
Mar 14 2017 23:11
i mean, it can be good and open
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:11
yeah it can be good
berceanu
@berceanu
Mar 14 2017 23:11
but what do they do with it
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:11
I think there is a big issue in its overrepresentation
berceanu
@berceanu
Mar 14 2017 23:11
you can use it in bioinformatics
for example
but people have been using it, for years
without the hype
now it's like, wait, we can use this for everything
you just need enough data
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:12
but if you play into the hype, you get million dollar grants from the NIH
if you don't play into the hype, then you're doing more mathy stuff... which means hundred thousand dollar grants from the NSF
Scott P. Jones
@ScottPJones
Mar 14 2017 23:13
The NIH should really give a big grant to improve a diffeq suite and a plotting suite that I know of ;-)
berceanu
@berceanu
Mar 14 2017 23:13
because people taking the decisions already bought into the hype
gnuplot?
:)
no no wait
xmgrame
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:13
There's just a lot of hype that neural nets and ML are the end-all be-all form of computation, and lots of money has shifted as though that is the case
berceanu
@berceanu
Mar 14 2017 23:13
xmgrace
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:13
while the rest of science... solves diffeqs
berceanu
@berceanu
Mar 14 2017 23:14
ye, all those losers the didn't discover deep learning yet
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:14
@ScottPJones actually, there is something along those lines... I got an on the spot postdoc offer today along with a project to push for an NIH grant.
I'll tell you about that sometime.
Scott P. Jones
@ScottPJones
Mar 14 2017 23:14
I've heard there's starting to be a bit of a backlash - CEOs believed all the promises that fresh out of a boot camp "data scientists" made, for doing the things like optimizing their ads for dishwashing soap ;-)
Great!
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:15
But yeah
berceanu
@berceanu
Mar 14 2017 23:15
that's the other problem
CEOs and people doing the hiring
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:15
my thing isn't that data science or deep learning is bad... it's just that it's not the answer to every problem like a lot of people think/say it is
berceanu
@berceanu
Mar 14 2017 23:15
they didn't even do the 4 week bootcamp
so if those guys are clueless, imagine them!
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:15
Deep learning tends to produce a black-box machine that can output predictions
Science wants to understand the mechanisms in the model
berceanu
@berceanu
Mar 14 2017 23:16
yup
Scott P. Jones
@ScottPJones
Mar 14 2017 23:16
Right! I have a lot of respect for the people doing ML/DL and data science (those that I've met at ODSC East and ODSC DataFest Boston)
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:16
I think that these tend to be pretty incompatible, though the tools of ML can be used with diffeq models
berceanu
@berceanu
Mar 14 2017 23:16
a black box with millions of tunable parameters, mind you
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:16
yeah
which is great if you want to predict correctly
Scott P. Jones
@ScottPJones
Mar 14 2017 23:16
but those are the people who know what they are doing
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:16
those things are really good at predicting
but prediction is different than understanding the mechanism
berceanu
@berceanu
Mar 14 2017 23:17
imagine newton's laws
but instead of just the mass
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:17
in fact, it's pretty easy to prove that the best predictive models are not the true models
berceanu
@berceanu
Mar 14 2017 23:17
you have one and a half million parameters
would you call that a good model
i mean, sure, maybe you can feed it 2 billion apples
and it gives you some force law
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:18
I think the biology example is more compelling
because there is no good model that we know of
that we can build it from
however, we have say a system of 200 genes
we want to know "how this system will behave"
(open-ended)
berceanu
@berceanu
Mar 14 2017 23:19
because biology is not derived from a small set of principles like physics
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:19
one camp is going along the machine learning path, doing deep learning, building systems that will predict protein concentrations
berceanu
@berceanu
Mar 14 2017 23:19
evolution has a million ways of achieving the same thing in different contexts
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:19
the other, is building diffeqs to understand how they interact
berceanu
@berceanu
Mar 14 2017 23:19
right
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:20
the ML group is getting predictions faster, but the ML methods don't tell you what's actually going on
Scott P. Jones
@ScottPJones
Mar 14 2017 23:20
Bloody amazing that all that complexity comes from such a simple alphabet
berceanu
@berceanu
Mar 14 2017 23:20
and the right camp, which is where all the money is, are optimizing some product recommendation system
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:20
in fact, they can't tell you anything about what proteins bind to each other
so both have a purpose, but one is not just growing but florishing in popularity and funding.
"It will predict when will get cancer"
but it will have no idea how to model what cancer is.
Scott P. Jones
@ScottPJones
Mar 14 2017 23:21
People like to follow fads too much :worried:
berceanu
@berceanu
Mar 14 2017 23:21
i guess people prefer results to understanding
if it tells me the cancer probabily i don't care how it did it
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:22
yeah
Scott P. Jones
@ScottPJones
Mar 14 2017 23:22
well, one that that kept coming up at the ds conferences, was that pure ML/DL approaches don't seem to work as well as hybrid approaches
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:22
yeah, I'm picking up some projects in that route
I mean, there has been so much good stuff in optimization and learning, it might as well be used in other places.
berceanu
@berceanu
Mar 14 2017 23:23
to me seing people so exicted about deep learning is like
imagine a hype about finite element methods
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:23
hahahaha
berceanu
@berceanu
Mar 14 2017 23:23
where everyone says they're like the bomb
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:24
"we make triangles, and like, fluids move man"
berceanu
@berceanu
Mar 14 2017 23:24
and will solve every problem known to man
:D
Scott P. Jones
@ScottPJones
Mar 14 2017 23:24
just stuffing a bunch of numbers into a black box and expecting it to pop up with 42 ;-)
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:24
At least one of the worst part of the DS craze is almost over
I call it "the clustering craze"
berceanu
@berceanu
Mar 14 2017 23:25
the worse part is once all this is over and people lose interest
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:25
deep learning craze is silly too, where everyone is just adding more and more hidden layers
but at least you get some kind of results out of it
prediction error decreases by 0.01%!
berceanu
@berceanu
Mar 14 2017 23:25
there will probably be some backlash
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:25
but new clustering algorithms?
berceanu
@berceanu
Mar 14 2017 23:25
in the funding
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:25
oh yeah, for sure
that will be the worst part
berceanu
@berceanu
Mar 14 2017 23:25
so even if you use it for good research in biology say
you suddenly find that funding dries up
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:26
"you guys spent all of this NIH money to do clustering? We should have just kept doing lab biology"
berceanu
@berceanu
Mar 14 2017 23:26
:D
Scott P. Jones
@ScottPJones
Mar 14 2017 23:26
Tom seemed to have some interesting ideas on DL, ways to do it better
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:26
I have only seen him talk about supervised methods, and yeah, his methods do seem interesting.
Scott P. Jones
@ScottPJones
Mar 14 2017 23:27
"We should have given it to Chris!" ;-)
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:27
hahaha nah, more like "we should have given it to systems biology!"
oh, actually, I need to clarify that now
berceanu
@berceanu
Mar 14 2017 23:27
i wonder if grocery stores will hire data scientists soon
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:27
because I found out that some places are using "systems biology" for "data science in bioinformatics" now
berceanu
@berceanu
Mar 14 2017 23:28
maybe it will be like, the accountant
a standard job that any company needs
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:28
grocery stores are never hiring again
more and more automation
berceanu
@berceanu
Mar 14 2017 23:28
systems biology seems interesting
like modelling all the complex networks
all the interactions
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:29
it is incredibly wide open
nobody knows what to do, so you can take any part of math/physics you like and start making relevant models and predictions out of it!
berceanu
@berceanu
Mar 14 2017 23:29
sounds more sound than an organ-based approach to medicine
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:29
it's super exiciting!
but... its deliverable results haven't been too much
berceanu
@berceanu
Mar 14 2017 23:29
sounds like
"it's good for publishing but not much else"
Scott P. Jones
@ScottPJones
Mar 14 2017 23:30
I'm pretty sure grocery store chains already do - have been doing it for decades, i.e. where to put things in which aisles, at what height, to get you to spend more money
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:30
biologists wanted the math to progress super fast in a span of (literally) 8 years
and weren't impressed
berceanu
@berceanu
Mar 14 2017 23:30
hahaha
we need this
when?
yesterday!
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:30
it's super dumb
I'm serious when I say this
berceanu
@berceanu
Mar 14 2017 23:31
but that's a nonlinear PDE you're trying to solve
who cares, i want an analytic solution and i want it now
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:31
The NIH created a whole set of "Centers for Systems Biology", multi-million dollar grants to get them propped up
Scott P. Jones
@ScottPJones
Mar 14 2017 23:31
Really?
berceanu
@berceanu
Mar 14 2017 23:31
so you're saying it's another fad?
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:31
UC Irvine was one of the first, along with Harvard, NC State, and a few other places
berceanu
@berceanu
Mar 14 2017 23:31
"nanotechonology"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:31
:smile: yeah, it was a fad
berceanu
@berceanu
Mar 14 2017 23:31
"systems biology"
what's next
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:32
but yeah, just a few years ago, they decided there weren't enough deliverables
berceanu
@berceanu
Mar 14 2017 23:32
and then they cut the funding?!
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:32
literally, there is no personalized medicine yet, so...
yeah, the centers are unfunded
berceanu
@berceanu
Mar 14 2017 23:32
like who's in charge there?
monkeys?!
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:32
well, they created a new setup where they are like
berceanu
@berceanu
Mar 14 2017 23:32
"you have 5 years to deliver personalized medicine"
"ok, funding's off"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:33
well, we'll fund some things for "developing a systems understanding of cancer"
so now, our whole center is putting in a whole new grant
and everyone is shifting focus to things which have the word cancer associated with it
because... the NIH said so.
I'll find out at the end of the month exactly how that's going: the Center for Complex Biological Systems (CCBS) has its retreat then
berceanu
@berceanu
Mar 14 2017 23:34
i know some people with chronic fatigue in the US, they told me funding is dismal
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:34
they'll give us a full slideshow of how the center is planning to get the next wave of funding.
berceanu
@berceanu
Mar 14 2017 23:34
so now i understand. it's not on the NIH hype list
hype keyword list
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:34
yeah
berceanu
@berceanu
Mar 14 2017 23:35
"diabetes. go!"
"we want quarterly reports"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:35
it's utterly bizzare that the research world has to live like this
berceanu
@berceanu
Mar 14 2017 23:35
it bizzare that it's ruled by people with no research experience
wait, what am i saying, research experience
look at trump
Scott P. Jones
@ScottPJones
Mar 14 2017 23:36
Well, our whole country is ruled by people with no experience - so, why should that be a surprise? ;-)
Yep, same thought at the same time
berceanu
@berceanu
Mar 14 2017 23:36
i mean
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:36
it's ruled by people who don't understand science and are self-centered
berceanu
@berceanu
Mar 14 2017 23:36
you think the guy knows what a matrix is?
or a cell
Scott P. Jones
@ScottPJones
Mar 14 2017 23:37
NPD all around
berceanu
@berceanu
Mar 14 2017 23:37
or a particle
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:37
I mean, there are literally separate funding paths for "exotic diseases". You know what classifies? Malaria.
berceanu
@berceanu
Mar 14 2017 23:37
ooh that sounds exotic
maybe cos people in "exotic" places get it
btw, i worked at a hedge fund for like a week
you know what surprised me most
Scott P. Jones
@ScottPJones
Mar 14 2017 23:39
Of course, for the next 4 years, don't expect any funding at all for science
berceanu
@berceanu
Mar 14 2017 23:39
the resources they had compared to academia
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:39
There was a chart someone showed me before. I really want to find it again. It was a chart of what diseases were super overrepresented in terms of funding vs the number of annual cases.
berceanu
@berceanu
Mar 14 2017 23:39
i mean, i was still doing my python stuff
but had like 5 programmers ready to implement everything in C++
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:40
Most highly represented: genetic and developemental diseases which are normally associated with Western Euorpean children.
Scott P. Jones
@ScottPJones
Mar 14 2017 23:40
Ufff
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:40
Then the presenter went on to show who related to which senator had a kid with that disease.
berceanu
@berceanu
Mar 14 2017 23:40
plus all the computing power you could wish for? want the newest Nvidia GPU? here you go
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:40
man... I should email for those slides
berceanu
@berceanu
Mar 14 2017 23:40
want 5 of them? no prob
no grant proposal to write
wish academia was like that
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:41
well, there is D.E. Shaw
Scott P. Jones
@ScottPJones
Mar 14 2017 23:41
Please do, I'd love to see that about the funding if you can find it
berceanu
@berceanu
Mar 14 2017 23:42
"my kid has a rare genetic disease"
"i think ill triple the funding"
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:42
it's an important national crisis!
berceanu
@berceanu
Mar 14 2017 23:42
just take from all the hepatitis pacients
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:42
Meanwhile, dengue fever research is essentially non-existant
berceanu
@berceanu
Mar 14 2017 23:42
fad-based culture
and the internet is making it worse
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:43
yup
that's the only way smartwatches could take off
berceanu
@berceanu
Mar 14 2017 23:44
i also want the slides if u get them
i think they'll kick us out soon :)
we were off topic like 2 hours ago
this is a whole different level
Scott P. Jones
@ScottPJones
Mar 14 2017 23:45
Hmmm, just looked up D.E. Shaw, "Senior Research Fellow at the Center for Computational Biology and Bioinformatics at Columbia University". I wish all investment (private and governmental) were being done by this sort of person ;-)
berceanu
@berceanu
Mar 14 2017 23:45
i totally forgot i'm in #symengine
Scott P. Jones
@ScottPJones
Mar 14 2017 23:46
Not the monkeys ;-)
Christopher Rackauckas
@ChrisRackauckas
Mar 14 2017 23:46
well, I found this really quickly
such efforts helped to increase new funding for rare-disease programmes by 3–15% each year at the US National Institutes of Health
for the record, total NIH funding has been stagnant over that period.
so...
let's take this over to the Julia channel :smile:
berceanu
@berceanu
Mar 14 2017 23:48
ahahaha
"they don't mind there"
"yesteday we shared recipes for cuban sandwitches"
Scott P. Jones
@ScottPJones
Mar 14 2017 23:48
Ah shucks, how could I have missed that!