-
Notifications
You must be signed in to change notification settings - Fork 0
/
analyze_actions.py
82 lines (73 loc) · 2.9 KB
/
analyze_actions.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
""" python ./mmlu_pro_analyze_actions.py ./output/batched/.../all.json, this will create a report in the same directory as the input file """
import json
import argparse
import os
import re
import csv
parser = argparse.ArgumentParser()
parser.add_argument("files", type=str, nargs="+")
args = parser.parse_args()
files = args.files
data = []
for file in files:
with open(file, "r", encoding='utf-8') as file:
data.append(json.load(file))
num_files = len(files)
if num_files == 1:
output_dir = os.path.dirname(files[0])
else:
output_dir = "./"
actions = [[] for _ in range(num_files)]
pattern = r"^action (\w+)."
action_types = []
for i in range(num_files):
for entry in data[i]:
action_for_entry = []
action_steps = len(entry["timestamp"]) - 1
valid_actions = len(entry["action"])
num_wait_actions = action_steps - valid_actions
completion_tokens = entry["usage"]["completion_tokens"]
assert completion_tokens.count(4) == num_wait_actions
start_pos = 0
for j in range(action_steps):
if completion_tokens[j] == 4:
action = "wait"
else:
matched = re.match(pattern, entry["action"][start_pos])
if matched:
action = matched.group(1)
else:
action = "unknown"
start_pos += 1
action_for_entry.append(action)
if action not in action_types:
action_types.append(action)
actions[i].append(action_for_entry)
actions_per_len = [{} for _ in range(num_files)]
actions_per_step = [{} for _ in range(num_files)]
for i in range(num_files):
for action_per_entry in actions[i]:
action_len = len(action_per_entry)
if action_len not in actions_per_len[i]:
actions_per_len[i][action_len] = []
actions_per_len[i][action_len].extend(action_per_entry)
for j in range(action_len):
if j+1 not in actions_per_step[i]:
actions_per_step[i][j+1] = []
actions_per_step[i][j+1].append(action_per_entry[j+1])
with open(os.path.join(output_dir, "actions_per_len.csv"), "w", encoding='utf-8') as file:
writer = csv.writer(file)
row_title = ["length"] + action_types
writer.writerow(row_title)
for i in range(num_files):
for length, actions in actions_per_len[i].items():
row = [length] + [actions.count(action)/len(actions) for action in action_types]
writer.writerow(row)
with open(os.path.join(output_dir, "actions_per_step.csv"), "w", encoding='utf-8') as file:
writer = csv.writer(file)
row_title = ["step"] + action_types
writer.writerow(row_title)
for i in range(num_files):
for step, actions in actions_per_step[i].items():
row = [step] + [actions.count(action)/len(actions) for action in action_types]
writer.writerow(row)