Advanced positional Indexing along dimensions

[ianis-g]

Hello everyone,

Is it possible to perform indexing instead of slicing along dimension using sel or isel functions ?

Thanks in advance for your help !

PS: Thank you very much for XDAS development ! :)

[atrabattoni]

Dear Ianis, Thank you for this very first technical question abour Xdas on github discussions !

You can definitevely perform index based selection with the Dataarray.isel method. For example:

da.isel(time=<something>)

This will check what is the axis corresponding to the time dimension. Let's imagine the time dimension is the first one in that case it will perform:

da[<something>]

if time is the second dimension it will perform this instead:

da[:, <something>]

and so on.

Now about the <something> part. What you can do or can't do depend on the type of coordinate an data objects the DataArray is composed of. In most cases it should work fine and all the type of indexing supported by Numpy should work (look at the numpy documentation on this). On that part, note that arr[1:2:3] is equivalent to arr[slice(1, 2, 3)]. It is the convinient form that python offers to express slices when indexing.

At the time of writing of this answer, coordinate objects should not be a problem. On the other side if you data array contain a virtual object, this could lead to some issue as this kind of object might not support advanced numpy indexing. In that case, the simplest work around is to first load the data to get a numpy data object and then to do your fancy indexing:

da = da.load()
da.isel(dim=<some_advanced_indexing>)

This is not optimal but should be efficient enough in most cases. And unfortunately Xdas provides obscure errors when you hit this kind of issues. Try to first make simple indexing on the virtual object, then load the minimal part of the data of interest and then make the advanced indexing.

Let me know if this answer you question !

[ianis-g]

I got it ! :)
Thanks !

[Linvill]

Hi all,

I'd like to ask a follow-up question here. Is it possible to use .isel to select multiple distance traces over the full time range of a DataArray, based on a list or something similar?
We’re working with DAS data from a single cable that was sequentially deployed in multiple boreholes. I’d like to exclude the traces that passed through air between the boreholes. In the attached figure, you can see an earthquake recorded in three boreholes, with the noisy air-coupled traces in between. Thanks in advance for the help!

[atrabattoni]

Hi @Linvill

If you want to do fancy selections while keeping virtual objects the more flexible way to go is to slice and concatenate.

You could do something like:

keep = [slice(<start1>, <end1>), ..., slice(<startN>, <endN>)]
objs = [da.sel(distance=slc) for slc in keep]
da_clean = xdas.concatenate(objs, dim="distance")

In this example I used sel so i would specifiy distance values but you can also use isel with distance indices.

Let me know if this works for you.

[Linvill]

Hi @atrabattoni,

thanks for your fast response, this works nicely, I even added the different objects of the different boreholes into a DataCollection, which works beautifully.

Thanks & Cheers