AttributeError: 'dict' object has no attribute 'shape' in varm

[gokceneraslan]

Following is an error I get when I read an anndata file which I unfortunately cannot share but I thought it might be helpful nevertheless:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-19-5c8f9dcd2470> in <module>
----> 1 hair_ad = sc.read('hair_follicle_and_glands.h5ad')

~/.miniconda/lib/python3.8/site-packages/scanpy/readwrite.py in read(filename, backed, sheet, ext, delimiter, first_column_names, backup_url, cache, cache_compression, **kwargs)
    110     filename = Path(filename)  # allow passing strings
    111     if is_valid_filename(filename):
--> 112         return _read(
    113             filename,
    114             backed=backed,

~/.miniconda/lib/python3.8/site-packages/scanpy/readwrite.py in _read(filename, backed, sheet, ext, delimiter, first_column_names, backup_url, cache, cache_compression, suppress_cache_warning, **kwargs)
    711     if ext in {'h5', 'h5ad'}:
    712         if sheet is None:
--> 713             return read_h5ad(filename, backed=backed)
    714         else:
    715             logg.debug(f'reading sheet {sheet} from file {filename}')

~/.miniconda/lib/python3.8/site-packages/anndata/_io/h5ad.py in read_h5ad(filename, backed, as_sparse, as_sparse_fmt, chunk_size)
    435     _clean_uns(d)  # backwards compat
    436 
--> 437     return AnnData(**d)
    438 
    439 

~/.miniconda/lib/python3.8/site-packages/anndata/_core/anndata.py in __init__(self, X, obs, var, uns, obsm, varm, layers, raw, dtype, shape, filename, filemode, asview, obsp, varp, oidx, vidx)
    306             self._init_as_view(X, oidx, vidx)
    307         else:
--> 308             self._init_as_actual(
    309                 X=X,
    310                 obs=obs,

~/.miniconda/lib/python3.8/site-packages/anndata/_core/anndata.py in _init_as_actual(self, X, obs, var, uns, obsm, varm, varp, obsp, raw, layers, dtype, shape, filename, filemode)
    509         # TODO: Think about consequences of making obsm a group in hdf
    510         self._obsm = AxisArrays(self, 0, vals=convert_to_dict(obsm))
--> 511         self._varm = AxisArrays(self, 1, vals=convert_to_dict(varm))
    512 
    513         self._obsp = PairwiseArrays(self, 0, vals=convert_to_dict(obsp))

~/.miniconda/lib/python3.8/site-packages/anndata/_core/aligned_mapping.py in __init__(self, parent, axis, vals)
    233         self._data = dict()
    234         if vals is not None:
--> 235             self.update(vals)
    236 
    237 

~/.miniconda/lib/python3.8/_collections_abc.py in update(self, other, **kwds)
    830         if isinstance(other, Mapping):
    831             for key in other:
--> 832                 self[key] = other[key]
    833         elif hasattr(other, "keys"):
    834             for key in other.keys():

~/.miniconda/lib/python3.8/site-packages/anndata/_core/aligned_mapping.py in __setitem__(self, key, value)
    149 
    150     def __setitem__(self, key: str, value: V):
--> 151         value = self._validate_value(value, key)
    152         self._data[key] = value
    153 

~/.miniconda/lib/python3.8/site-packages/anndata/_core/aligned_mapping.py in _validate_value(self, val, key)
    213                 f"value.index does not match parent’s axis {self.axes[0]} names"
    214             )
--> 215         return super()._validate_value(val, key)
    216 
    217     @property

~/.miniconda/lib/python3.8/site-packages/anndata/_core/aligned_mapping.py in _validate_value(self, val, key)
     48         """Raises an error if value is invalid"""
     49         for i, axis in enumerate(self.axes):
---> 50             if self.parent.shape[axis] != val.shape[i]:
     51                 right_shape = tuple(self.parent.shape[a] for a in self.axes)
     52                 raise ValueError(

AttributeError: 'dict' object has no attribute 'shape'

[gokceneraslan]

OK, figured out having a key with "/" character in varm e.g. adata.varm['a/b'] = pd.DataFrame() causes the error.

Code to reproduce:

import scanpy as sc

ad = sc.datasets.paul15()
ad.varm['a/b'] = ad.X.T
ad.write('ppp.h5ad') # interestingly, success
ad = sc.read('ppp.h5ad') #uninterestingly, failure

[gokceneraslan]

Apparently it's already known (#321 (comment)), so closing this one, damn it already ate up 30min of my life :/

[ivirshup]

It would be good to at least warn users about this when we can.