Interview with Felix Cremer
This is a modal window.
Das Video konnte nicht geladen werden, da entweder ein Server- oder Netzwerkfehler auftrat oder das Format nicht unterstützt wird.
Formale Metadaten
| Titel |
| |
| Serientitel | ||
| Anzahl der Teile | 44 | |
| Autor | ||
| Lizenz | CC-Namensnennung 3.0 Deutschland: Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen. | |
| Identifikatoren | 10.5446/63424 (DOI) | |
| Herausgeber | ||
| Erscheinungsjahr | ||
| Sprache | ||
| Produzent | ||
| Produktionsjahr | 2023 | |
| Produktionsort | Wageningen |
Inhaltliche Metadaten
| Fachgebiet | ||
| Genre | ||
| Abstract |
| |
| Schlagwörter |
8
9
10
11
15
17
19
22
23
25
27
29
32
36
37
41
00:00
KurvenanpassungDimensionsanalyseNeuroinformatikSymboltabelleStrategisches SpielParallele SchnittstelleMultiplikationsoperatorDistributionenraumTaskFormale SpracheGraphPhysikalisches SystemMittelwertFunktionalBesprechung/Interview
02:12
MultiplikationsoperatorFunktionalCASE <Informatik>MomentenproblemSatellitensystemAnalysisDigitalisierungMereologieService providerRelativitätstheorieTypentheorieWürfelBitSurjektivitätZeitreihenanalyseSkriptspracheFlächeninhaltFaltungsoperatorNeuronales NetzAbstandOrtsoperatorVariableDifferenteBesprechung/Interview
05:09
Computeranimation
Transkript: Englisch(automatisch erzeugt)
00:00
My name is Felix Klima, I'm working at the Max Planck Institute for Biogeochemistry, and I presented about how to use the Julia language and the YAXAways.jl package and the
00:24
Julia DataQ's ecosystem to handle large roster data. And the idea is that you want to have an efficient system to access your data, but you also want to provide functionality to not think
00:44
too much about the dimensions and where, how the dimensions are ordered, but you want to provide like, you want to compute along certain axis, for example, you want to compute the average of time,
01:03
and then you don't want to need to know whether the time dimension is the first dimension in the array or whether it's the second or third, and therefore you want to be able to access these data with symbols so that you don't have to think about these things.
01:26
So one advantage for YAXAways compared to your XRA is that for certain computational tasks, YAXAways has a much easier distribution or parallelization strategy. So in XRA, it depends on the task and you need to
01:47
compute or set up the whole computational graph for all chunks, and in XRA, this is not, in the XRA, that is not necessary, so you need to, you don't need to set up the whole graph
02:02
for the parallelization, and therefore you can have certain computations that are possible in XRA that are not possible in XRA. I think there's a very small learning curve to get started because you can easily, as I showed, use your Python or R functionality, so if you have a function that you,
02:25
that works in Python or R on a time series, and then you want to just use this function along every time series in your latitude, longitude, time data cube, you can easily use YAXAways to parallelize this, and then you can then think about switching other parts of your
02:52
analysis onto Julia, but I think for a first start, you can just use your Python script inside of the YAXAways package.
03:08
I think data cubes are going to stay, so I think there are two things that I see that are currently happening which are kind of related but a bit distinct is that, one is that up to now a lot of
03:24
data cubes was someone had to provide your data cube and had to make sure that every variable is on the same grid, and so you have this difference between data cube provider and data cube users,
03:41
and I think there's more and more movement into combining data cubes that are on different grids so that you can just more easily combine different satellites or satellite data with your analysis data or with other types of data, and the other area that we explore also is going from the
04:09
longitude latitude grid to digital global grid systems, so the idea is that on the longitude
04:21
latitude grid you have the problem that higher latitudes have a smaller area, so you have to adjust for that and depending on the functions you are applying that is not that easy and the advantage of a DGGS is that you have always the same area always at least roughly the same
04:43
neighboring distances and then it's easier to apply for example a convolutional neural networks because the convolution is the same at all positions of the globe which is not the case at the moment or not the case with longitude and latitude grids.