|
From: | Felix Höfling |
Subject: | Re: [h5md-user] proposal 100 - update |
Date: | Tue, 16 Jun 2015 11:13:16 +0200 |
User-agent: | Opera Mail/12.16 (Linux) |
Hi all, On Mon, Jun 01, 2015 at 10:01:06PM +0200, Pierre de Buyl wrote:On Mon, Jun 01, 2015 at 03:54:16PM -0400, Peter Colberg wrote: > On Fri, May 29, 2015 at 08:46:56AM +0200, Pierre de Buyl wrote:> > But, in another likely usage, imagine that step is not linear and that> > time[i]=step[i]*DT for all values of i. > >> > My "new" extension is that if you have a non-scalar dataset for "step" and a> > scalar dataset for "time", you can compute the value in time as > > time[i]=step[i]*time > > If time depends on step, I would suggest this scheme: > > \-- step[Inf] > | +-- timestep: Float[] > \-- value[Inf, …] > > However, to keep things simple, I would not favour this extension. If > one already goes through the trouble of writing the variable-length > step dataset, it is not much more work to write the time dataset using > the same datatype and dataspace.It is more about writing less data than convenience per se. I don't see this asa "must have" though.Any other opinion on this?As a reminder, I propose that step[Inf] and time[] (that is, a scalar dataset) allows one to evaluate the time at index i as (step[i]+step.offset*time[]).P
I think the correct formula reads: time[i] = time.offset + (step[i] - step.offset) * time[]The benefit of such an extension to the proposal is really small, so I would drop it. I think that most situations are covered by the following to scenarios:
i) for "rare" sampling (only at a few steps) it is most straightforward to write step and time explicitly, as it is in 1.0.
ii) for more frequent, regular sampling, the proposal offers the calculation of the sampling time from the array index, the increment (scalar dataset "time") and the offset (attribute time.offset)
Pierre, is their a specific application which motivates your suggestion? Felix
[Prev in Thread] | Current Thread | [Next in Thread] |