On Monday, 15 June 2015 at 19:35:16 UTC, Laeeth Isharc wrote:
On Monday, 15 June 2015 at 17:28:52 UTC, John Colvin wrote:
On Monday, 15 June 2015 at 17:11:26 UTC, Laeeth Isharc wrote:
On Monday, 15 June 2015 at 06:51:44 UTC, John Colvin wrote:
[...]
Yes - I had noticed same, but don't yet have the experience
(and don't have available time for now) to do much about it.
Was looking at Facebook torch to see how that fitted with
bokeh and inotebook, but glad you actually did something to
make it happen.
I haven't looked at torch much, is it up to the hype?
To be direct: I don't know, as I haven't used it. I was just
interested in seeing how it fitted with ipython and bokeh in
particular.
I liked your talk at dconf.
Thank you :)
Looks like progress is being made on planting the seeds of a
matrix implementation and computation library on top.
Ilya Yaroshenko is doing great work here.
Having a way to use D code to explore data within an approach
of rapid iteration will be useful too, and I think your work is
a great start on this (seems like its value as it is is very
much greater than you modestly suggest, and also I am amazed by
how short the code is).
Well, as i briefly mentioned in my talk, it's really very easy to
do, Jupyter/IPython makes it simple. Of course, behind the scenes
Pyd is doing the hard work.
At the moment I am using a combination of the python interface
to Bokeh and D code to do the work. It would be great to port
the bokeh server side stuff to D - I have made a start, but I
haven't the space for a concerted effort. The funny thing is
there is nothing to the server side - all you do is have a
shallow layer that translates the cumulative state from API
calls to JSON; the hard work of drawing stuff is done in
Javascript on the client.
Whether it's Bokeh or not, having some kind of visualisation
solution that fits well with iteratively exploring data in D
will be nice. (What you have done means it's already enough to
be very usable, but there may be benefits to moving the
charting itself to D).
Visualisation is one of those things that no-one seems to have
got right, despite the huge efforts that have gone in to it. I
have ideas about D-based visualisation and already my own basic
3-D plotting library for huge datasets, but for now the link to
python is key to get quick and easy access to the latest tools
that the scientific and finance world are creating.
Perhaps not for scientific applications, but for finance:
reporting is also something that is an important and rising
theme in use cases. Not just for accounting, but for example
to understand portfolio risk in a richer way. You have for a
long time been able to hook up reportlab with D code using PyD,
but this element of rapid iteration was not so easy - small
frictions have large consequences when you don't have a big
team.
So in any case, I really appreciate your work.
One more thing - any thoughts on best spreadsheet like
interface either purely on browser or within ipython to display
results and allow user to enter parameters? My friend, Giles
Thomas, wrote Resolver but that's too big and aimed at
something different, and I don't know if it is being maintained
properly.
There is http://nbviewer.ipython.org/gist/rossant/9463955 (see
https://www.youtube.com/watch?v=_JZFSFR6Yeo for demo) which looks
like a good starting point, but it's very basic.
If https://github.com/quantopian/qgrid supported editing (perhaps
it does already, but I can't see any mention of it) then that
would probably be better.