individual scores on the four languages on which the system performed best. If less than four
scores are supplied, the global score is NaN.
scores = []
for lang in ["en", "de", "es", "it", "fa"]:
scores.append(spearman_evaluate(vectors, read_semeval_monolingual(lang)))
top_scores = sorted(scores, key=lambda x: x["acc"] if not np.isnan(x["acc"]) else 0)[-4:]
acc_average = tmean([score["acc"] for score in top_scores])
low_average = tmean([score["low"] for score in top_scores])
high_average = tmean([score["high"] for score in top_scores])
return pd.Series(
[acc_average, low_average, high_average],