Sparse array is not open for reading when writing

This code complains for array not being open for reading (previous step).

The array definition is:

   domain = tiledb.Domain(tiledb.Dim(name="cols", domain=(978307200, 4102358400), 
                                                tile=1000, dtype=np.int64))
    schema = tiledb.ArraySchema(domain=domain, sparse=True,
                        attrs=[tiledb.Attr(name="ndvi", dtype=np.float32),
                               tiledb.Attr(name="kcb",dtype=np.float32),
                               tiledb.Attr(name="nn",dtype=np.float32),
                               tiledb.Attr(name="eto",dtype=np.float32),    
                               tiledb.Attr(name="trigger",dtype=np.dtype('b'))])

`

What I want to do is, for a given timestamp (dimension cols), write a list of attributes (vegatation indexes).

And when computed those values, the code that tries to write today’s recordings is:

   #TiledDB writing
    with tiledb.SparseArray(grupo_parcela, mode='w') as parcelaAttrsW:                      
                        parcelaAttrsW[tsHoy]['trigger'] = trigger
                        parcelaAttrsW[tsHoy]['eto']=etoParcela 
 

So, direct access by indexing for writing seems to need previous read, and thus makes opening ‘w’ fail. Should I construct a numpy array first and then write it to file, should it be something like:

                      parcelaAttrsW[tsHoy]=full array numpy

In my opinion, RW mode is something quite useful that has been requested before:

https://github.com/TileDB-Inc/TileDB-Py/issues/58

Kind regards!

[stavrospapadopoulos]

Here is a more detailed example when you have two attributes a and b in a sparse array with a single dimension t:

with tiledb.SparseArray(array_name, mode='w') as A:
    # These are the timestamps you would like to write at: 1, 5, 6
    t = [1, 5, 6]
    # These are the values for attribute "a" and "b" at timestamps 1, 5, 6, respectively
    a = np.array(([1, 2, 3]));
    b = np.array(([3, 3, 10]));
    # This is how you write
    A[t] = {"a":a, "b":b}

The above will create the following sparse cells in A (recall, t is the dimension, and a and b are the attributes):

t=1, a=1, b=3
t=5, a=2, b=3
t=6, a=3, b=10

My comment above meant to clarify that whenever you write some cells to the array, you need to provide values for all attributes. In other words, in the above example, A[t] = {"a":a} or A[t] = {"b":b} or A[t] = a or A[t] = b would error out - you need to provide values for both a and b.

[stavrospapadopoulos]

I think there are two different issues here.

1. Writing multiple attributes to an array. Currently, when writing to an array, you need to provide all attributes as a dictionary of numpy arrays. You can find more details here. Therefore, you need:

parcelaAttrsW[tsHoy] = {"trigger": trigger, "eto": etoParcela, ... <all othe attributes here>}

1. Read-Write mode. Currently, we support opening an array object either in read or write mode. However, TileDB allows you to open the same array in both read and write mode, but using two different array objects (one with mode ='r' and one with mode='w'). This decision was taken in order to simplify the consistency semantics; when opening an array for reading, TileDB “views” only writes that happened at or before the timestamp the array is getting opened at. If we mix reads and writes in the same object, these semantics can become confusing. We continue to consider the read-write mode though and hope to add it in the future with some clear semantics.

Could you please send us the exact error message you are getting for (1) above, as I think it is confusing and we could improve it.