-
Notifications
You must be signed in to change notification settings - Fork 5
/
evaluation.py
206 lines (177 loc) · 7.53 KB
/
evaluation.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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
import os
import yaml
import json
import argparse
import pandas as pd
from utils.parser import parse_code_to_workflow, parse_markdown_to_workflow
from utils.comfy import execute_workflow
with open('./config.yaml', 'r') as file:
config = yaml.load(file, Loader=yaml.FullLoader)
proxy_config = config['proxy']
os.environ['http_proxy'] = proxy_config['http_proxy']
os.environ['https_proxy'] = proxy_config['https_proxy']
def main(args):
with open('./dataset/query/meta.json', 'r') as file:
metadata = yaml.load(file, Loader=yaml.FullLoader)
record = dict()
for agent_name in args.agent_name:
record[agent_name] = dict()
for task_id, task_info in metadata.items():
record[agent_name][f'task_{task_id}'] = {
'task_info': task_info,
'num_runs': 0,
'num_passes_1': 0,
'num_passes_2': 0
}
for agent_name in args.agent_name:
print(f'[Evaluation] agent {agent_name}')
for task_id in metadata.keys():
prefix = f'{args.save_path}/{agent_name}/task_{task_id}'
print(f'[Evaluation] task {task_id}')
for run_id in os.listdir(prefix):
record[agent_name][f'task_{task_id}']['num_runs'] += 1
checkpoint = os.path.join(prefix, run_id)
print(f'[Evaluation] checkpoint {checkpoint}')
# Skip: already evaluated
output_path = os.path.join(checkpoint, 'output')
if os.path.exists(output_path):
record[agent_name][f'task_{task_id}']['num_passes_1'] += 1
record[agent_name][f'task_{task_id}']['num_passes_2'] += 1
print('skipped: already evaluated')
continue
# Check: pipeline error
log_path = os.path.join(checkpoint, 'run.log')
with open(log_path, 'r', errors='ignore') as file:
log = file.read()
if 'Failed to generate workflow' in log or 'Failed to refine workflow' in log:
print('skipped: pipeline error')
continue
# Case: standard representation
if agent_name in [
'zero_shot_agent',
'few_shot_agent',
'cot_agent',
'rag_agent',
'gen_agent',
'code_gen_agent',
'single_gen_agent'
]:
# Check: no file
code_path = os.path.join(checkpoint, 'code.py')
if not os.path.exists(code_path):
print('skipped: no file')
continue
# Check: empty code
with open(code_path, 'r') as file:
code = file.read()
if code.strip() == '':
print('skipped: empty code')
continue
# Check: invalid workflow
try:
workflow = parse_code_to_workflow(code)
except Exception as error:
print('skipped: invalid workflow')
continue
# Case: json representation
elif agent_name in [
'json_gen_agent'
]:
# Check: no file
json_path = os.path.join(checkpoint, 'workflow.json')
if not os.path.exists(json_path):
print('skipped: no file')
continue
# Check: invalid format
try:
with open(json_path, 'r') as file:
workflow = json.load(file)
except Exception as error:
print('skipped: invalid workflow')
continue
# Case: list representation
elif agent_name in [
'list_gen_agent',
]:
# Check: no file
markdown_path = os.path.join(checkpoint, 'markdown.md')
if not os.path.exists(markdown_path):
print('skipped: no file')
continue
# Check: empty list
with open(markdown_path, 'r') as file:
markdown = file.read()
if markdown.strip() == '':
print('skipped: empty list')
continue
# Check: invalid workflow
try:
workflow = parse_markdown_to_workflow(markdown)
except Exception as error:
print('skipped: invalid workflow')
continue
# Record: pass 1
record[agent_name][f'task_{task_id}']['num_passes_1'] += 1
# Check: execution failure
try:
status, outputs = execute_workflow(workflow)
except Exception as error:
print('skipped: execution failure')
continue
# Check: invalid status
if status['status_str'] != 'success':
print('skipped: invalid status')
continue
# Check: empty output
if len(outputs) == 0:
print('skipped: empty output')
continue
# Save: execution output
output_path = os.path.join(checkpoint, 'output')
os.makedirs(output_path, exist_ok=True)
for file_name, output in outputs.items():
file_path = os.path.join(output_path, file_name)
with open(file_path, 'wb') as file:
file.write(output)
# Record: pass 2
record[agent_name][f'task_{task_id}']['num_passes_2'] += 1
summary = {
'Agent Name': [],
'(Run Level) Pass Rate 1': [],
'(Run Level) Pass Rate 2': [],
'(Task Level) Pass Rate 1': [],
'(Task Level) Pass Rate 2': []
}
for agent_name, agent_record in record.items():
num_runs, num_tasks = 0, len(agent_record)
run_passes_1, run_passes_2 = 0, 0
task_passes_1, task_passes_2 = 0, 0
for task_record in agent_record.values():
num_runs += task_record['num_runs']
run_passes_1 += task_record['num_passes_1']
run_passes_2 += task_record['num_passes_2']
if task_record['num_passes_1'] > 0:
task_passes_1 += 1
if task_record['num_passes_2'] > 0:
task_passes_2 += 1
summary['Agent Name'].append(agent_name)
summary['(Run Level) Pass Rate 1'].append(run_passes_1 / num_runs)
summary['(Run Level) Pass Rate 2'].append(run_passes_2 / num_runs)
summary['(Task Level) Pass Rate 1'].append(task_passes_1 / num_tasks)
summary['(Task Level) Pass Rate 2'].append(task_passes_2 / num_tasks)
summary = pd.DataFrame(summary)
print(summary.to_string())
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument(
'--agent_name',
nargs='+',
type=str
)
parser.add_argument(
'--save_path',
default='./checkpoint/benchmark',
type=str
)
args = parser.parse_args()
main(args)