评价推荐系统时最重要的两个问题是评分准确率和Top-N推荐问题。评分准确率问题常用的度量标准有Root Mean Squared Error （RMSE），Precision，Recall等。Top-N推荐常用的度量标准有NDCG，Average Reciprocal Hit-Rank（ARHR）等。
参考代码：
import numpy as np
def DCG(A, test_set):
# ------ 计算 DCG ------ #
dcg = 0
for i in range(len(A)):
# 给r_i赋值，若r_i在测试集中则为1，否则为0
r_i = 0
if A[i] in test_set:
r_i = 1
dcg += (2 ** r_i - 1) / np.log2((i + 1) + 1) # (i+1)是因为下标从0开始
return dcg
def IDCG(A, test_set):
# ------ 将在测试中的a排到前面去，然后再计算DCG ------ #
A_temp_1 = [] # 临时A，用于存储r_i为1的a
A_temp_0 = [] # 临时A，用于存储r_i为0的a
for a in A:
if a in test_set:
# 若a在测试集中则追加到A_temp_1中
A_temp_1.append(a)
else:
# 若a不在测试集中则追加到A_temp_0中
A_temp_0.append(a)
A_temp_1.extend(A_temp_0)
idcg = DCG(A_temp_1, test_set)
return idcg
def NDCG(A, test_set):
dcg = DCG(A, test_set) # 计算DCG
idcg = IDCG(A, test_set) # 计算IDCG
if dcg == 0 or idcg == 0:
ndcg = 0
else:
ndcg = dcg / idcg
return ndcg
参考资料：https://www.136.la/nginx/show-174149.html
import numpy as np
def mean_reciprocal_rank(rs):
"""Score is reciprocal of the rank of the first relevant item
First element is 'rank 1'. Relevance is binary (nonzero is relevant).
Example from http://en.wikipedia.org/wiki/Mean_reciprocal_rank
>>> rs = [[0, 0, 1], [0, 1, 0], [1, 0, 0]]
>>> mean_reciprocal_rank(rs)
0.61111111111111105
>>> rs = np.array([[0, 0, 0], [0, 1, 0], [1, 0, 0]])
>>> mean_reciprocal_rank(rs)
0.5
>>> rs = [[0, 0, 0, 1], [1, 0, 0], [1, 0, 0]]
>>> mean_reciprocal_rank(rs)
0.75
Args:
rs: Iterator of relevance scores (list or numpy) in rank order
(first element is the first item)
Returns:
Mean reciprocal rank
"""
rs = (np.asarray(r).nonzero()[0] for r in rs)
return np.mean([1. / (r[0] + 1) if r.size else 0. for r in rs])
def r_precision(r):
"""Score is precision after all relevant documents have been retrieved
Relevance is binary (nonzero is relevant).
>>> r = [0, 0, 1]
>>> r_precision(r)
0.33333333333333331
>>> r = [0, 1, 0]
>>> r_precision(r)
0.5
>>> r = [1, 0, 0]
>>> r_precision(r)
1.0
Args:
r: Relevance scores (list or numpy) in rank order
(first element is the first item)
Returns:
R Precision
"""
r = np.asarray(r) != 0
z = r.nonzero()[0]
if not z.size:
return 0.
return np.mean(r[:z[-1] + 1])
def precision_at_k(r, k):
"""Score is precision @ k
Relevance is binary (nonzero is relevant).
>>> r = [0, 0, 1]
>>> precision_at_k(r, 1)
0.0
>>> precision_at_k(r, 2)
0.0
>>> precision_at_k(r, 3)
0.33333333333333331
>>> precision_at_k(r, 4)
Traceback (most recent call last):
File "", line 1, in ?
ValueError: Relevance score length < k
Args:
r: Relevance scores (list or numpy) in rank order
(first element is the first item)
Returns:
Precision @ k
Raises:
ValueError: len(r) must be >= k
"""
assert k >= 1
r = np.asarray(r)[:k] != 0
if r.size != k:
raise ValueError('Relevance score length < k')
return np.mean(r)
def average_precision(r):
"""Score is average precision (area under PR curve)
Relevance is binary (nonzero is relevant).
>>> r = [1, 1, 0, 1, 0, 1, 0, 0, 0, 1]
>>> delta_r = 1. / sum(r)
>>> sum([sum(r[:x + 1]) / (x + 1.) * delta_r for x, y in enumerate(r) if y])
0.7833333333333333
>>> average_precision(r)
0.78333333333333333
Args:
r: Relevance scores (list or numpy) in rank order
(first element is the first item)
Returns:
Average precision
"""
r = np.asarray(r) != 0
out = [precision_at_k(r, k + 1) for k in range(r.size) if r[k]]
if not out:
return 0.
return np.mean(out)
def mean_average_precision(rs):
"""Score is mean average precision
Relevance is binary (nonzero is relevant).
>>> rs = [[1, 1, 0, 1, 0, 1, 0, 0, 0, 1]]
>>> mean_average_precision(rs)
0.78333333333333333
>>> rs = [[1, 1, 0, 1, 0, 1, 0, 0, 0, 1], [0]]
>>> mean_average_precision(rs)
0.39166666666666666
Args:
rs: Iterator of relevance scores (list or numpy) in rank order
(first element is the first item)
Returns:
Mean average precision
"""
return np.mean([average_precision(r) for r in rs])
def dcg_at_k(r, k, method=0):
"""Score is discounted cumulative gain (dcg)
Relevance is positive real values. Can use binary
as the previous methods.
Example from
http://www.stanford.edu/class/cs276/handouts/EvaluationNew-handout-6-per.pdf
>>> r = [3, 2, 3, 0, 0, 1, 2, 2, 3, 0]
>>> dcg_at_k(r, 1)
3.0
>>> dcg_at_k(r, 1, method=1)
3.0
>>> dcg_at_k(r, 2)
5.0
>>> dcg_at_k(r, 2, method=1)
4.2618595071429155
>>> dcg_at_k(r, 10)
9.6051177391888114
>>> dcg_at_k(r, 11)
9.6051177391888114
Args:
r: Relevance scores (list or numpy) in rank order
(first element is the first item)
k: Number of results to consider
method: If 0 then weights are [1.0, 1.0, 0.6309, 0.5, 0.4307, ...]
If 1 then weights are [1.0, 0.6309, 0.5, 0.4307, ...]
Returns:
Discounted cumulative gain
"""
r = np.asfarray(r)[:k]
if r.size:
if method == 0:
return r[0] + np.sum(r[1:] / np.log2(np.arange(2, r.size + 1)))
elif method == 1:
return np.sum(r / np.log2(np.arange(2, r.size + 2)))
else:
raise ValueError('method must be 0 or 1.')
return 0.
def ndcg_at_k(r, k, method=0):
"""Score is normalized discounted cumulative gain (ndcg)
Relevance is positive real values. Can use binary
as the previous methods.
Example from
http://www.stanford.edu/class/cs276/handouts/EvaluationNew-handout-6-per.pdf
>>> r = [3, 2, 3, 0, 0, 1, 2, 2, 3, 0]
>>> ndcg_at_k(r, 1)
1.0
>>> r = [2, 1, 2, 0]
>>> ndcg_at_k(r, 4)
0.9203032077642922
>>> ndcg_at_k(r, 4, method=1)
0.96519546960144276
>>> ndcg_at_k([0], 1)
0.0
>>> ndcg_at_k([1], 2)
1.0
Args:
r: Relevance scores (list or numpy) in rank order
(first element is the first item)
k: Number of results to consider
method: If 0 then weights are [1.0, 1.0, 0.6309, 0.5, 0.4307, ...]
If 1 then weights are [1.0, 0.6309, 0.5, 0.4307, ...]
Returns:
Normalized discounted cumulative gain
"""
dcg_max = dcg_at_k(sorted(r, reverse=True), k, method)
if not dcg_max:
return 0.
return dcg_at_k(r, k, method) / dcg_max参考资料：Ranking Metrics · GitHubRanking Metrics. GitHub Gist: instantly share code, notes, and snippets.
https://gist.github.com/bwhite/3726239
