Concatenate unique numpy arrays with counters

Question

Is there a simple yet efficient way to concatenate two unique numpy arrays which have counters?

Example:

values1 = np.array(['host1', 'host2', 'host3', 'host6'])
counts1 = np.array([2,5,2,4])

values2 = np.array(['host3', 'host1', 'host4'])
counts2 = np.array([5,7,1])

I'd like to have a result like:

values_res = np.array(['host1', 'host2', 'host3', 'host6', 'host4'])
counts_res = np.array([9,5,7,4,1])

They do not need to be ordered, but values_res does need to be unique.

I could iterate over the elements in the array, but that would not be efficient. I'd like to use vectorization somehow.

Answer 1

This is probably faster (specially for larger arrays) and is ordered:

values_res, idx = np.unique(np.hstack((values1, values2)), return_inverse=True)
counts_res = np.bincount(idx, np.hstack((counts1, counts2)))

output:

['host1' 'host2' 'host3' 'host4' 'host6']
[9. 5. 7. 1. 4.]

---

Comparison using benchit:

#@Ehsan's solution
def m1(values1, values2, counts1, counts2):
  values_res, idx = np.unique(np.hstack((values1, values2)), return_inverse=True)
  counts_res = np.bincount(idx, np.hstack((counts1, counts2)))
  return values_res, counts_res

#@Chris's solution
def m2(values1, values2, counts1, counts2):

  values_res, counts_res = zip(*dict(Counter(dict(zip(values1,counts1))) + Counter(dict(zip(values2,counts2)))).items())
  return values_res, counts_res


in_ = {n:[np.random.choice(values_res, n), np.random.choice(values_res, n), np.random.randint(1,100,n), np.random.randint(1,100,n)] for n in [10,100,1000,10000]}

output:

m1 is faster in this setting