BUG: sklearn/utils/multiclass.type_of_target inconsistent for similar inputs

Description

sklearn/utils/multiclass.type_of_target(y) returns different results for practically identical ‘y’ values when y a list and when y is an numpy array.

Steps/Code to Reproduce

from sklearn.utils import multiclass
import numpy as np
x = [[1, 1], [0, 1]]
y = np.asarray(x)
a = multiclass.type_of_target(x)
b = multiclass.type_of_target(y) 
print(a, b)
assert a == b

Expected Results

‘multilabel-indicator, multilabel-indicator’ No assertion error

Actual Results

‘multiclass-multioutput, multilabel-indicator’ Throws assertion error

AssertionError

System Info

System:
    python: 3.6.7rc2 (v3.6.7rc2:4893861ab5, Oct 13 2018, 05:25:29)  [GCC 4.2.1 Compatible Apple LLVM 6.0 (clang-600.0.57)]
executable: /Users/maxwellaladago/Documents/pub/sklearn-dev/venv/bin/python
   machine: Darwin-18.7.0-x86_64-i386-64bit
Python deps:
       pip: 19.2.2
setuptools: 39.1.0
   sklearn: 0.22.dev0
     numpy: 1.17.0
     scipy: 1.3.1
    Cython: 0.29.13
    pandas: None
matplotlib: None
    joblib: 0.13.2

Will you consider a PR for this? Also, can you clarify what is_multilable's behaviour should be? The documentation says it Check if y is in a multilabel format but it seems to be checking whether y is a multi_label_indicator except that y is not restricted to be either 0 or 1.

[jnothman]

In the sense that we always report the strictest matching type, yes it is correct.

[cmarmo]

Running the code above with the current version throws no error. If this is a duplicate of #7931 it has been fixed by #14865. I’m closing it.