"""Calculate evaluation metrics such as hits@k."""
from typing import Any, Dict, List, Union
import numpy as np
def _hits_from_ndarray(nn_ind, gold, k, hits_counter):
for hits_at_k in k:
for i, _ in enumerate(nn_ind):
if i in gold and gold[i] in nn_ind[i][:hits_at_k]:
hits_counter[hits_at_k] += 1
return hits_counter
def _hits_from_dict(nn_ind, gold, k, hits_counter):
for hits_at_k in k:
for i, v in nn_ind.items():
if i in gold and gold[i] in v[:hits_at_k]:
hits_counter[hits_at_k] += 1
return hits_counter
[docs]def hits(
nn_ind: Union[np.ndarray, Dict[Any, List]],
gold: Dict[Any, Any], # source -> target
k=None,
) -> Dict[int, float]:
"""Show hits@k.
Parameters
----------
nn_ind: array or dict
Contains the indices of the nearest neighbors for source entities.
gold: dict
Map of source indices to the respective target indices
k: array
Which ks should be evaluated
Returns
-------
hits: dict
k: relative hits@k values
Examples
--------
>>> from kiez.evaluate import hits
>>> import numpy as np
>>> nn_ind = np.array([[1, 2, 3], [2, 3, 4], [3, 4, 5], [4, 5, 6]])
>>> gold = {0: 2, 1: 4, 2: 3, 3: 4}
>>> hits(nn_ind, gold)
{1: 0.5, 5: 1.0, 10: 1.0}
"""
if k is None:
k = [1, 5, 10]
k.sort()
hits_counter = {hits_at_k: 0 for hits_at_k in k}
if isinstance(nn_ind, (np.ndarray, list)):
hits_counter = _hits_from_ndarray(nn_ind, gold, k, hits_counter)
elif isinstance(nn_ind, Dict):
hits_counter = _hits_from_dict(nn_ind, gold, k, hits_counter)
return {hits_at_k: k_val / len(gold) for hits_at_k, k_val in hits_counter.items()}
