-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbuild_results_table.py
178 lines (141 loc) · 6.51 KB
/
build_results_table.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
import argparse
from collections import defaultdict
from pathlib import Path
import natsort
import numpy as np
from natsort import natsorted
from sklearn.metrics import average_precision_score
from scipy.stats import rankdata
import pandas as pd
from tabulate import tabulate
from tqdm import tqdm
def get_metrics(df, score):
score_arr = df[score].values
score_arr[(score_arr == np.inf) ] = score_arr[score_arr != np.inf].max() + 1
score_arr[np.isnan(score_arr)] = score_arr[~np.isnan(score_arr)].max() + 1
result = {
# "mean_rank": rankdata(-score_arr)[df["has_problem"]].mean(),
"ap": average_precision_score(df["has_problem"], score_arr) * 100,
}
return result
def get_results_rows(df):
ordered_columns = ['random', 'ppl', 'p_mean', 'p_min', 'aum']
df_to_save = df.pivot_table(index='score', columns='problem', values='ap')
df_to_save = df_to_save.reindex(ordered_columns)
problem_to_scores = {}
for problem in df_to_save.keys().unique():
problem_to_scores[problem] = df_to_save[problem].values
return problem_to_scores
def build_results_df(df):
results_df = defaultdict(list)
for problem in sorted(df["problem"].unique()) + ["all"]:
if problem.startswith("none") or problem == "unknown":
continue
if problem != "all":
df_problem = df[(df["problem"] == problem) | (df["problem"] == "none_" + problem)].copy()
else:
df_problem = df[(df["problem"] != "unknown")].copy()
df_problem["has_problem"] = df_problem["has_problem"] > 0
results_df["ap"].append(df_problem["has_problem"].mean() * 100)
results_df["problem"].append(problem)
results_df["score"].append("random")
results = get_metrics(df_problem, "ppl")
for k, v in results.items():
results_df[k].append(v)
results_df["problem"].append(problem)
results_df["score"].append("ppl")
results = get_metrics(df_problem, "aum")
for k, v in results.items():
results_df[k].append(v)
results_df["problem"].append(problem)
results_df["score"].append("aum")
results = get_metrics(df_problem, "p_mean")
for k, v in results.items():
results_df[k].append(v)
results_df["problem"].append(problem)
results_df["score"].append("p_mean")
results = get_metrics(df_problem, "p_min")
for k, v in results.items():
results_df[k].append(v)
results_df["problem"].append(problem)
results_df["score"].append("p_min")
return pd.DataFrame(results_df)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--data", required=True)
parser.add_argument("--single", action="store_true")
parser.add_argument("--adapted", action="store_true")
parser.add_argument("--task_agg", default=None)
args = parser.parse_args()
if args.adapted:
scores_dir = Path("adapted_scores")
else:
scores_dir = Path("scores")
size_to_lines = {"small": [], "base": [], "large": [], "xl": []}
sizes = ["small", "base", "large", "xl"]
seeds = [42, 43, 44]
for size in tqdm(list(size_to_lines)):
results_rows = defaultdict(list)
for seed in tqdm(seeds):
files = natsorted(scores_dir.glob(f"p4_runs_large_{args.data}_{seed}_t5-{size}-lm-adapt_checkpoint-*.csv"))
if args.single:
files = [files[-1]]
if not files:
print(f"No files found for seed {seed} and size {size}")
print("Exiting...")
exit(1)
all_scores = {"ppl": [], "p_mean": [], "p_min": [], "aum": []}
for checkpoint in tqdm(files):
df = pd.read_csv(checkpoint).fillna("")
df = df[df["problem"] != "unknown"]
all_scores["ppl"].append(df["ppl"])
all_scores["p_mean"].append(-df["p_mean"])
all_scores["p_min"].append(-df["p_min"])
all_scores["aum"].append(df["aum"])
for score_name, scores in all_scores.items():
scores = [np.array(score).astype(float) for score in scores]
df[score_name] = np.mean(scores, axis=0)
df["has_problem"] = ~df["problem"].str.startswith("none")
dataset_to_problem = {}
for dataset in df["dataset"].unique():
df_dataset = df[df["dataset"] == dataset]
problem = [i for i in df_dataset["problem"].unique() if i != "none"]
if problem:
dataset_to_problem[dataset] = problem[0]
else:
dataset_to_problem[dataset] = "none"
if args.task_agg == "mean":
df = df.groupby("dataset").agg({"ppl": "mean", "aum": "mean", "p_mean": "mean", "p_min": "mean", "has_problem": "mean", "problem": "first"})
df["problem"] = [dataset_to_problem[dataset] for dataset in df.index]
elif args.task_agg == "median":
df = df.groupby("dataset").agg({"ppl": "median", "aum": "median", "p_mean": "median", "p_min": "median", "has_problem": "mean", "problem": "first"})
df["problem"] = [dataset_to_problem[dataset] for dataset in df.index]
results_df = build_results_df(df)
problem_to_scores = get_results_rows(results_df)
for problem, scores in problem_to_scores.items():
results_rows[problem].append(scores)
all_lines = [] # should be XX.X±X.X
for problem in sorted(results_rows.keys()):
result_strings = []
mean = np.mean(results_rows[problem], axis=0)
std = np.std(results_rows[problem], axis=0)
for m, s in zip(mean, std):
result_strings.append(f"${m:.1f}_" + "\\textit{" + f"{s:.1f}" + "}$")
all_lines.append("\t".join([problem] + result_strings))
size_to_lines[size].extend(all_lines)
all_lines = []
for i in range(len(list(size_to_lines.values())[0])):
line = size_to_lines[sizes[0]][i].split("\t")[:2]
for size in sizes:
line.extend(size_to_lines[size][i].split("\t")[2:])
all_lines.append("\t".join(line))
results_name = f"results/{args.data}"
if args.adapted:
results_name += "_adapted"
if args.task_agg:
results_name += f"_{args.task_agg}"
if args.single:
results_name += "_single"
results_name += ".tsv"
with open(results_name, "w") as f:
f.write("\n".join(all_lines))