-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathaggregate_results.py
190 lines (169 loc) · 11.9 KB
/
aggregate_results.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
from functools import reduce
from pathlib import Path
import seaborn as sns
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.lines as lines
from autorank._util import test_normality, rank_multiple_nonparametric
from scipy.stats import pearsonr
from file_checker import get_evaluation_sets
from static import *
def test_statistical_significance(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches):
mask_performance = {
evaluation_set: {metric: {recommender: [] for recommender in recommenders} for metric in metrics} for
evaluation_set in get_evaluation_sets()}
base_path = f"./{AGGREGATION_FOLDER}"
Path(base_path).mkdir(parents=True, exist_ok=True)
with open(f"{base_path}/statistical_significance_test.txt", "w") as results_file:
for recommender in recommenders:
for metric in metrics:
for evaluation_set in get_evaluation_sets():
data_set_mask_performance = []
for idx, data_set_name in enumerate(data_set_names):
merged_leaderboard = pd.read_csv(f"./{DATA_FOLDER}/{data_set_name}/"
f"{EVALUATION_FOLDER}_{recommender}_"
f"{metric}_{topn_score}_{topn_sample}_{num_batches}/"
f"{LEADERBOARD_FILE}", header=0, sep=",")
merged_leaderboard = merged_leaderboard.drop(columns=[f"score_{evaluation_set}"])
merged_leaderboard.rename(
columns={"score_validation": f"score_validation_{idx}", "score_test": f"score_test_{idx}"},
inplace=True)
data_set_mask_performance.append(merged_leaderboard)
mask_performance[evaluation_set][metric][recommender] = reduce(
lambda left, right: pd.merge(left, right, on='mask'), data_set_mask_performance)
stat_data = (mask_performance[evaluation_set][metric][recommender].iloc[:, 1:].T * -1).reset_index(
drop=True)
rank_data = pd.DataFrame(stat_data.values, columns=list(stat_data))
alpha_normality = 0.05 / len(rank_data.columns)
print("-" * 80, file=results_file)
print(f"Statistical test for {evaluation_set} {metric} {recommender}.", file=results_file)
all_normal, pvals_shapiro = test_normality(rank_data, alpha_normality, False)
friedman_nemenyi = rank_multiple_nonparametric(rank_data, 0.05, True, all_normal, "ascending", None,
force_mode=None)
# result = RankResult(friedman_nemenyi.rankdf, friedman_nemenyi.pvalue, friedman_nemenyi.cd,
# friedman_nemenyi.omnibus, friedman_nemenyi.posthoc, all_normal, pvals_shapiro,
# None, None, None, 0.05, alpha_normality, len(rank_data), None, None,
# None, None, friedman_nemenyi.effect_size)
# if result.pvalue >= result.alpha:
# raise ValueError(f"The test result is not significant (p={result.pvalue}).")
mean_rank_of_topn_mask = friedman_nemenyi.rankdf.loc[rank_data.shape[1] - 1]["meanrank"]
number_of_masks_with_same_difference = abs(
friedman_nemenyi.rankdf["meanrank"] - mean_rank_of_topn_mask) <= friedman_nemenyi.cd
percentage_of_masks_with_same_difference = sum(
number_of_masks_with_same_difference / len(friedman_nemenyi.rankdf))
print(f"Percentage of masks without statistically significant difference: "
f"{percentage_of_masks_with_same_difference}", file=results_file)
def plot_aggregated_data_topn_performance(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches):
# retrieve recommender performance in terms of the best other selection versus top-n and store in dictionary
recommender_performance = {
evaluation_set: {metric: {recommender: [] for recommender in recommenders} for metric in metrics} for
evaluation_set in get_evaluation_sets()}
base_path = f"./{AGGREGATION_FOLDER}"
Path(base_path).mkdir(parents=True, exist_ok=True)
for data_set_name in data_set_names:
for recommender in recommenders:
for metric in metrics:
merged_leaderboard = pd.read_csv(f"./{DATA_FOLDER}/{data_set_name}/"
f"{EVALUATION_FOLDER}_{recommender}_"
f"{metric}_{topn_score}_{topn_sample}_{num_batches}/"
f"{LEADERBOARD_FILE}", header=0, sep=",")
sorted_by_mask_index = merged_leaderboard.sort_values(by="mask", ascending=False)
for evaluation_set in get_evaluation_sets():
topn_mask_performance = sorted_by_mask_index.iloc[0][f"score_{evaluation_set}"]
sorted_by_score = merged_leaderboard.sort_values(by=f"score_{evaluation_set}", ascending=False)
best_other_mask_validation_performance = \
sorted_by_score.drop(index=sorted_by_mask_index.index[0]).iloc[0][f"score_{evaluation_set}"]
recommender_performance[evaluation_set][metric][recommender].append(
(best_other_mask_validation_performance / topn_mask_performance) - 1)
data_set_names_list = [data_set_name for data_set_name in data_set_names]
# plot the performance of the best other selection strategy compared to top-n selection strategy
for evaluation_set in get_evaluation_sets():
for metric in metrics:
for plot_type in ["results", "baselines"]:
performance_df = pd.DataFrame(recommender_performance[evaluation_set][metric])
performance_df.rename(columns={"implicit-mf": "Implicit MF",
"user-knn": "User-based kNN",
"item-knn": "Item-based kNN",
"bayesian-personalized-ranking": "Bayesian Personalized Ranking",
"alternating-least-squares": "Alternating Least Squares",
"logistic-mf": "Logistic MF",
"item-item-cosine": "Item-based kNN Cosine Sim.",
"item-item-tfidf": "Item-based kNN TF-IDF Sim.",
"item-item-bm25": "Item-based kNN BM25 Sim.",
"random": "Random",
"popularity": "Popularity"}, inplace=True)
performance_df.index = data_set_names_list
metric_name = "nDCG" if metric == "ndcg" else "Precision"
plot_df = None
if plot_type == "results":
plt.figure(figsize=(12, 4))
plot_df = performance_df.drop(columns=["Random", "Popularity"])
elif plot_type == "baselines":
plt.figure(figsize=(12, 1))
if len(performance_df) > 3:
plot_df = performance_df[["Popularity", "Random"]]
else:
plot_df = performance_df[["Random", "Popularity"]]
sns_plot_df = pd.melt(plot_df.reset_index(), id_vars="index")
optimum_line = plt.axvline(x=0, color='r', linestyle='-', label="a")
number_hues = len(plot_df)
if number_hues == 6:
markers = ["^", "X", "o", "p", "D", "*"]
elif number_hues == 5:
markers = ["^", "X", "o", "p", "D"]
elif number_hues == 4:
markers = ["^", "X", "o", "p"]
elif number_hues == 3:
markers = ["^", "X", "o"]
if 3 <= number_hues <= 6:
sc_axis = sns.scatterplot(data=sns_plot_df, x="value", y="variable", hue="index", s=75,
palette="colorblind", style="index", markers=markers, legend="full")
if plot_type == "baselines":
plt.ylim(-0.5, 1.5)
handles, _ = sc_axis.get_legend_handles_labels()
plt.legend(handles=handles,
labels=[f"Identical relative {metric_name} performance"] + data_set_names_list,
bbox_to_anchor=(0., 1.02, 1., .102), loc='lower left',
ncol=number_hues + 1)
else:
sns.boxplot(data=sns_plot_df, x="value", y="variable", orient="h",
palette="colorblind", showfliers=False, boxprops=dict(alpha=.8))
sns.stripplot(data=sns_plot_df, x="value", y="variable", hue="index", orient="h",
palette="dark:black", s=5)
legend_proxy = lines.Line2D([], [], color='black', marker='o', linestyle='None', markersize=5)
plt.legend(handles=[optimum_line, legend_proxy],
labels=[f"Identical relative {metric_name} performance",
f"Relative {metric_name} performance of one data set"],
bbox_to_anchor=(0., 1.02, 1., .102), loc='lower left', ncol=2)
plt.xlabel(f"Relative {metric_name} performance")
plt.ylabel("")
plt.savefig(f'{base_path}/shopping_{evaluation_set}_{metric}_top-n_performance_{plot_type}.pdf',
format="pdf", bbox_inches='tight')
plt.clf()
plt.cla()
plt.close()
def calculate_generalization_score(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches):
recommender_correlation = {metric: {recommender: [] for recommender in recommenders} for metric in metrics}
for data_set_name in data_set_names:
for recommender in recommenders:
for metric in metrics:
merged_leaderboard = pd.read_csv(f"./{DATA_FOLDER}/{data_set_name}/"
f"{EVALUATION_FOLDER}_{recommender}_"
f"{metric}_{topn_score}_{topn_sample}_{num_batches}/"
f"{LEADERBOARD_FILE}", header=0, sep=",")
recommender_correlation[metric][recommender].append(
pearsonr(merged_leaderboard["score_test"], merged_leaderboard["score_validation"])[0])
base_path = f"./{AGGREGATION_FOLDER}"
Path(base_path).mkdir(parents=True, exist_ok=True)
with open(f"{base_path}/generalization_score.txt", "w") as results_file:
for recommender in recommenders:
for metric in metrics:
recommender_correlation[metric][recommender] = np.mean(recommender_correlation[metric][recommender])
print(f"Pearson correlation for {metric} - {recommender}: "
f"{recommender_correlation[metric][recommender]}", file=results_file)
def aggregate_results(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches):
test_statistical_significance(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches)
plot_aggregated_data_topn_performance(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches)
calculate_generalization_score(data_set_names, recommenders, metrics, topn_score, topn_sample, num_batches)
return