Is there any function equivalent to np.unique for generic object in Python

Question

np.unique() can return indices of first occurrence, indices to reconstruct, and occurrence count. Is there any function/library that can do the same for any Python object?

Answer 1

Not as such. You can get similar functionality using different classes depending on your needs.

unique with no extra flags has a similar result to set:

unique_value = set(x)

collections.Counter simulates return_counts:

counts = collections.Counter(x)
unique_values = list(counts.keys())
unique_counts = list(counts.values())

To mimic return_index, use list.index on a set or Counter. This assumes that the container is a list

first_indices = [x.index(k) for k in counts]

To simulate return_inverse, we look at how unique is actually implemented. unique sorts the input to get the runs of elements. A similar technique can be acheived via sorted (or in-place list.sort) and itertools.groupby:

s = sorted(zip(x, itertools.count()))
inverse = [0] * len(x)
for i, (k, g) in enumerate(itertools.groupby(s, operator.itemgetter(0))):
    for v in g:
        inverse[v[1]] = i

In fact, the groupby approach encodes all the options:

s = sorted(zip(x, itertools.count()))
unique_values = []
first_indices = []
unique_counts = []
inverse = [0] * len(x)
for i, (k, g) in enumerate(itertools.groupby(s, operator.itemgetter(0))):
    unique_values.append(k)
    count = 1
    v = next(g)
    inverse[v[1]] = i
    first_indices.append(v[0])
    for v in g:
        inverse[v[1]] = i
        count += 1
    unique_counts.append(count)