scipy.stats.mode does not accept non-numeric values

[xinyu-dev]

In Tutorial_Reference_mappying.ipynb, the following lines raise an error

for k in idx_list:
    if faiss_imported:
        idx = labels[k]
    else:
        idx, sim = get_similar_vectors(test_emebd[k][np.newaxis, ...], ref_cell_embeddings, k)
    pred = mode(ref_embed_adata.obs[cell_type_key][idx], axis=0) # I made this change. scipy.stats.mode no longer accepts non-numeric values
    preds.append(pred[0][0])

TypeError Traceback (most recent call last)
/tmp/ipykernel_21913/2423575529.py in ?()
17 if faiss_imported:
18 idx = labels[k]
19 else:
20 idx, sim = get_similar_vectors(test_emebd[k][np.newaxis, ...], ref_cell_embeddings, k)
---> 21 pred = (ref_embed_adata.obs[cell_type_key][idx], axis=0) # I made this change. scipy.stats.mode no longer accepts non-numeric values
22 preds.append(pred[0][0])
23 # preds.append(stat_mode(np.array(ref_embed_adata.obs[cell_type_key][idx])))
24

/usr/local/lib/python3.10/dist-packages/scipy/stats/_axis_nan_policy.py in ?(failed resolving arguments)
519 # behavior of those would break backward compatibility.
520
521 if sentinel:
522 samples = _remove_sentinel(samples, paired, sentinel)
--> 523 res = hypotest_fun_out(*samples, **kwds)
524 res = result_to_tuple(res)
525 res = _add_reduced_axes(res, reduced_axes, keepdims)
526 return tuple_to_result(*res)

/usr/local/lib/python3.10/dist-packages/scipy/stats/_stats_py.py in ?(a, axis, nan_policy, keepdims)
507 message = ("Argument a is not recognized as numeric. "
508 "Support for input that cannot be coerced to a numeric "
509 "array was deprecated in SciPy 1.9.0 and removed in SciPy "
510 "1.11.0. Please consider np.unique.")
--> 511 raise TypeError(message)
512
513 if a.size == 0:
514 NaN = _get_nan(a)

TypeError: Argument a is not recognized as numeric. Support for input that cannot be coerced to a numeric array was deprecated in SciPy 1.9.0 and removed in SciPy 1.11.0. Please consider np.unique.mode

My current workaround:

from statistics import mode as stat_mode

for k in idx_list:
    if faiss_imported:
        idx = labels[k]
    else:
        idx, sim = get_similar_vectors(test_emebd[k][np.newaxis, ...], ref_cell_embeddings, k)
    # pred = mode(ref_embed_adata.obs[cell_type_key][idx], axis=0) # I made this change. scipy.stats.mode no longer accepts non-numeric values
    # preds.append(pred[0][0])
    preds.append(stat_mode(np.array(ref_embed_adata.obs[cell_type_key][idx])))

Just want to confirm if this is workaround is appropriate. Thanks!

[subercui]

Thank you @xinyu-dev ! and yes, looks like it is related to scipy version changes. The workaround looks good to me. I am also going to update a fix right now. Going to use the pandas built-in value counts, it can be faster supposedly