can you save an object array using zarr?

Question

Following zarr's tutorial, I'm trying to save a list of list of ints to a persistent zarr:

• Failed method 1:

  import numcodecs, zarr
  zarr.save("path/to/zarr", [[1], [2]], dtype=object, object_codec=numcodecs.JSON())
  

• Failed method 2:

  import numcodec, zarr
  z = zarr.array([[1], [2]], dtype=object, object_codec=numcodecs.JSON())
  zarr.save("path/to/zarr", z, dtype=object, object_codec=numcodecs.JSON())
  

Both methods output ValueError: missing object_codec for object array

Answer 1

I question why you wouldn't just save your list of ints as an array in Zarr. However, it is generally possible to write object dtypes to zarr arrays:

import numpy as np
import zarr
from numcodecs import JSON

# slightly contrived example of an object array
x = np.empty(1, dtype=object)
x[0] = ['foo', 2, 'bar']

za = zarr.open_array('path/to/zarr', shape=x.shape, dtype=object, object_codec=JSON())

print(zarr.open_array(store)[:][0])  # --> ['foo', 2, 'bar']