comparing_rankings.py

import measures
import random
import string
import numpy as np
np.random.seed(0)

n_digits = 3
n_catalog = 50
n_ranking = 10
k = 20

# create a catalog of items and a reference and predicted ranking
catalog = [''.join(random.choices(string.ascii_uppercase + string.digits, k=n_digits)) for _ in range(n_catalog)]
reference_ranking = np.random.choice(catalog,n_ranking,replace=False)
predicted_ranking = np.random.choice(catalog,n_ranking,replace=False)

# find the ground truth relevance of the items in the predicted ranking
ground_truth_relevance = np.array([int(predicted_item in reference_ranking) for predicted_item in predicted_ranking])
predicted_relevance = np.ones(n_ranking,dtype=int)

true_order = np.argsort(-ground_truth_relevance)

reference = [ground_truth_relevance[x] for x in true_order]
hypothesis = [predicted_relevance[x] for x in true_order] # equals to predicted_relevance

# evaluate
## precision at k
my_precision = len(set(reference_ranking[:k])&set(predicted_ranking[:k]))/k
## precision
precision_k = measures.find_precision_k(reference,hypothesis,k)
## ndcg
ndcg = measures.find_ndcg(reference,hypothesis)
## rankdcg
rankdcg = measures.find_rankdcg(reference,hypothesis)
# rankdcg = measures.find_rankdcg(reference,np.random.permutation(reference))

print(reference)
print(hypothesis)

print('my_precision',my_precision)
print('precision_k',precision_k)
print('ndcg',ndcg)
print('rankndcg',rankdcg)