Datasets:

ArXiv:
File size: 15,973 Bytes
9f3bc09
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
import asyncio
import json
import os
import argparse
from typing import Dict, List, Optional
from tqdm import tqdm
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.messages import TextMessage
from autogen_agentchat.ui import Console
from autogen_core import CancellationToken
from autogen_ext.models.openai import AzureOpenAIChatCompletionClient
from autogen_core.models import ModelFamily
from prompts import thought_prompt, sys_prompt, identity_prompt
from tools import *
from utils import extract_json, identity_mapping_dict, setup_logging_and_config, process_qa, rag_url_dict
from datetime import datetime

# max steps for the agent to generate the answer
MAX_STEPS = 20


class AgentResponse:
    """Class to represent agent response structure."""
    def __init__(self, thoughts: str, response: str):
        self.thoughts = thoughts
        self.response = response


def parse_arguments() -> argparse.Namespace:
    """Parse command line arguments."""
    parser = argparse.ArgumentParser(
        description="Run EgoLife QA agent with chain-of-thought reasoning"
    )
    
    # Model configuration
    parser.add_argument(
        "--model", 
        type=str, 
        default="gpt-4.1",
        help="Model to use for the agent"
    )
    parser.add_argument(
        "--api_version", 
        type=str, 
        default="2024-09-01-preview",
        help="API version for the model"
    )
    
    # Data configuration
    parser.add_argument(
        "--data_path", 
        type=str, 
        default="./egor1-bench/QA-egolife/",
        help="Path to the data directory"
    )
    parser.add_argument(
        "--identity", 
        type=str, 
        default="A1",
        help="Identity to use for the agent"
    )
    
    # Output configuration
    parser.add_argument(
        "--result_dir", 
        type=str, 
        default="./results",
        help="Directory to save results"
    )
    parser.add_argument(
        "--log_dir", 
        type=str, 
        default="./logs",
        help="Directory to save logs"
    )
    parser.add_argument(
        "--cache_dir", 
        type=str, 
        default="./cache",
        help="Directory for caching"
    )
    
    # Processing options
    parser.add_argument(
        "--explicit_answer", 
        action="store_true",
        help="Use explicit answer termination"
    )
    parser.add_argument(
        "--observation_type", 
        type=str, 
        default="all_actions", 
        choices=["single", "all", "all_actions", "null"],
        help="Type of observation to include in prompts"
    )
    
    # Resume and specific data options
    parser.add_argument(
        "--resume", 
        action="store_true",
        help="Resume processing from error files"
    )
    parser.add_argument(
        "--gen_specific_data", 
        action="store_true",
        help="Process only specific data IDs"
    )
    parser.add_argument(
        "--specific_data_path", 
        type=str, 
        default="./data_statistics/error_list_results_aobs_gpt-41_A1.txt",
        help="Path to specific data list file (.txt or .json)"
    )
    
    return parser.parse_args()


def load_json_data(file_path: str) -> List[Dict]:
    """Load JSON data from file with error handling."""
    try:
        with open(file_path, "r", encoding="utf-8") as f:
            return json.load(f)
    except (json.JSONDecodeError, FileNotFoundError, UnicodeDecodeError) as e:
        raise ValueError(f"Failed to load {file_path}: {e}")


def load_specific_data_ids(file_path: str) -> List[int]:
    """Load specific data IDs from a text or JSON file."""
    if file_path.endswith(".txt"):
        with open(file_path, "r", encoding="utf-8") as f:
            data_str = f.readlines()[0].strip()
            # Handle both comma-separated and bracket-enclosed formats
            data_str = data_str.strip("[]")
            return [int(x.strip()) for x in data_str.split(",") if x.strip()]
    
    elif file_path.endswith(".json"):
        error_data = load_json_data(file_path)
        return [int(item["ID"]) for item in error_data]
    
    else:
        raise ValueError(f"Unsupported file format: {file_path}")


def load_resume_data_ids(identity: str) -> List[int]:
    """Load data IDs that need to be resumed from error files."""
    error_ids = []
    
    # Load NA errors
    na_error_path = f"data_gen/errors/error_list_na_{identity}.json"
    if os.path.exists(na_error_path):
        try:
            na_errors = load_json_data(na_error_path)
            error_ids.extend([int(d["ID"]) for d in na_errors])
        except Exception as e:
            print(f"Warning: Could not load NA errors from {na_error_path}: {e}")
    
    # Load no-answer errors (incomplete processing)
    no_answer_path = f"data_gen/errors/error_list_no_answer_{identity}.json"
    if os.path.exists(no_answer_path):
        try:
            no_answer_errors = load_json_data(no_answer_path)
            error_ids.extend([
                int(error_d["ID"]) 
                for error_d in no_answer_errors 
                if len(error_d.get("cot", [])) < MAX_STEPS
            ])
        except Exception as e:
            print(f"Warning: Could not load no-answer errors from {no_answer_path}: {e}")
    
    return list(set(error_ids))  # Remove duplicates


def filter_data_by_ids(data: List[Dict], target_ids: List[int]) -> List[Dict]:
    """Filter data to only include items with IDs in target_ids."""
    return [item for item in data if item["ID"] in target_ids]


def setup_model_client(args: argparse.Namespace) -> AzureOpenAIChatCompletionClient:
    """Set up the model client based on arguments."""
    endpoint, deployment, subscription_key = setup_logging_and_config(args.model)
    
    if args.model == "gpt-4.1":
        return AzureOpenAIChatCompletionClient(
            azure_deployment=deployment,
            azure_endpoint=endpoint,
            model="gpt-41",
            api_version="2025-01-01-preview",
            api_key=subscription_key,
            model_info={
                "family": ModelFamily.GPT_41,
                "function_calling": True,
                "json_output": True,
                "structured_output": True,
                "vision": False,
            }
        )
    else:
        return AzureOpenAIChatCompletionClient(
            azure_deployment=deployment,
            azure_endpoint=endpoint,
            model=args.model,
            api_version=args.api_version,
            api_key=subscription_key,
        )


def get_tools(args: argparse.Namespace) -> List:
    """Get the appropriate tools based on arguments."""
    if args.explicit_answer:
        return [rag, video_llm, vlm, terminate_explicit]
    else:
        return [rag, video_llm, vlm, terminate]


def get_system_prompt(args: argparse.Namespace) -> str:
    """Get the appropriate system prompt based on version."""
    return sys_prompt + f"\n{identity_prompt.format(identity=f'{args.identity}_{identity_mapping_dict[args.identity]}')}"


def create_resume_prompt(dp: Dict, base_prompt: List[TextMessage]) -> List[TextMessage]:
    """Create a prompt for resuming from previous CoT steps."""
    prompt = base_prompt.copy()
    
    if dp.get("cot"):
        prompt.append(TextMessage(
            source="assistant", 
            content=f'Previous observations: {dp["cot"]}'
        ))
        prompt.append(TextMessage(
            source="user",
            content=(
                "Now you are given the previous actions and observations you have made before, "
                "continue to try your best to answer the question using different tools. "
                f"You must provide an answer to the question before step {MAX_STEPS}."
            )
        ))
    
    return prompt


def update_prompt_with_observation(
    prompt: List[TextMessage], 
    observation: Dict, 
    observation_type: str
) -> List[TextMessage]:
    """Update prompt with new observation based on observation type."""
    if observation_type == "single":
        # Replace with single latest observation
        return [
            prompt[0], 
            TextMessage(source="assistant", content=f'Previous observations: {observation}')
        ]
    elif observation_type == "all":
        # Append all previous observations
        prompt.append(TextMessage(
            source="assistant", 
            content=f'Previous observations (step-{observation["step"]}): {observation}'
        ))
    elif observation_type == "all_actions":
        # Append all previous actions
        prompt.append(TextMessage(
            source="assistant", 
            content=f'Previous actions (step-{observation["step"]}): {observation["tool"]}'
        ))
    elif observation_type == "null":
        # Don't append any observation
        pass
    
    return prompt


async def process_single_qa(
    qa: Dict, 
    agent: AssistantAgent, 
    args: argparse.Namespace,
    result_dir: str
) -> Optional[Dict]:
    """Process a single QA item with the agent."""
    dp = process_qa(qa, args.explicit_answer)
    dp_path = os.path.join(result_dir, f"{dp['ID']}.json")
    
    # Handle existing files
    if os.path.exists(dp_path) and not args.resume:
        print(f"Overwriting {dp['ID']}")
    
    if args.resume and os.path.exists(dp_path):
        print(f"Resuming {dp['ID']}")
        with open(dp_path, "r", encoding="utf-8") as f:
            dp = json.load(f)
    
    # Create initial prompt
    base_prompt = [TextMessage(
        content=dp["question"] + "\n\n" + thought_prompt, 
        source="user"
    )]
    
    # Set up for resume if needed
    if args.resume and dp.get("cot") and len(dp["cot"]) > 0:
        step = len(dp["cot"]) - 1
        print(f"Resuming from {dp['ID']} at step {step}")
        dp["cot"] = dp["cot"][:-1]  # Remove last incomplete step
        prompt = create_resume_prompt(dp, base_prompt)
    else:
        step = 0
        prompt = base_prompt
    
    # Main processing loop
    while step < MAX_STEPS:
        print(f"Step: {step}")
        try:
            result = await Console(agent.run_stream(
                task=prompt, 
                cancellation_token=CancellationToken()
            ))
        except Exception as e:
            print(f"Error at step {step}: {e}")
            return {
                "id": dp["ID"],
                "prompt": [msg.content for msg in prompt],
                "error": str(e),
                "step": step
            }
        
        step += 1
        messages = result.messages
        
        # Extract information from messages
        thought = None
        tool_call = None
        tool_summary = None
        
        for message in messages:
            if message.type == "ThoughtEvent":
                thought = message.content
            elif message.type == "ToolCallRequestEvent":
                tool_call = message.content[0]
            elif message.type == "ToolCallSummaryMessage":
                tool_summary = message.content
        
        # Handle termination
        if tool_call and "terminate" in tool_call.name.lower():
            observation = {
                "step": step,
                "thought": thought,
                "answer": extract_json(tool_call.arguments)["answer"]
            }
            dp['cot'].append(observation)
            break
        
        # Handle regular tool usage
        else:
            observation = {
                "step": step,
                "thought": thought,
                "tool": {
                    "id": tool_call.id if tool_call else None,
                    "name": tool_call.name if tool_call else None,
                    "arguments": tool_call.arguments if tool_call else None
                },
                "observation": tool_summary
            }
            
            dp['cot'].append(observation)
            prompt = update_prompt_with_observation(prompt, observation, args.observation_type)
    
    # Save results
    with open(dp_path, "w", encoding="utf-8") as f:
        json.dump(dp, f, indent=4)
    
    return None  # No error


async def main() -> None:
    """Main function to orchestrate the QA processing."""
    args = parse_arguments()
    
    # Set up environment
    os.environ["LOG_DIR"] = args.log_dir
    if os.environ.get("IDENTITY") is None:
        os.environ["IDENTITY"] = args.identity
    if os.environ.get("RAG_URL") is None:
        os.environ["RAG_URL"] = rag_url_dict[args.identity]
        assert os.environ["RAG_URL"] is not None, "RAG_URL is not set"
    cache_dir = os.environ.get("CACHE_DIR", args.cache_dir)
    os.makedirs(args.result_dir, exist_ok=True)
    os.makedirs(cache_dir, exist_ok=True)
    os.makedirs(args.log_dir, exist_ok=True)
    
    # Load data based on mode
    data_path = os.path.join(
        args.data_path, 
        f"EgoLifeQA_{args.identity}_{identity_mapping_dict[args.identity]}.json"
    )
    
    print(f"Current identity: {args.identity}")
    
    try:
        if args.gen_specific_data:
            print("Loading specific data...")
            # Use specific_data_path or fall back to specific_data_list for compatibility
            specific_path = args.specific_data_path
            if not os.path.exists(specific_path) and hasattr(args, 'specific_data_list'):
                specific_path = args.specific_data_list
            
            if os.path.exists(specific_path):
                target_ids = load_specific_data_ids(specific_path)
            else:
                # Fallback to default path format
                default_path = f"./data_statistics/error_list_{args.identity}.txt"
                target_ids = load_specific_data_ids(default_path)
            
            all_data = load_json_data(data_path)
            egolife_qa_data = filter_data_by_ids(all_data, target_ids)
            print(f"Loaded {len(egolife_qa_data)} specific items")
            
        elif args.resume:
            print("Loading resume data...")
            target_ids = load_resume_data_ids(args.identity)
            all_data = load_json_data(data_path)
            egolife_qa_data = filter_data_by_ids(all_data, target_ids)
            print(f"Loaded {len(egolife_qa_data)} items for resume")
            
        else:
            print(f"Loading all data from: {data_path}")
            egolife_qa_data = load_json_data(data_path)
            print(f"Loaded {len(egolife_qa_data)} items")
            
    except Exception as e:
        print(f"Error loading data: {e}")
        return
    
    if not egolife_qa_data:
        print("No data to process.")
        return
    
    # Set up model and agent components
    model_client = setup_model_client(args)
    tools = get_tools(args)
    sys_prompt_text = get_system_prompt(args)
    
    # Process data
    errors = []
    
    for qa in tqdm(egolife_qa_data, desc="Processing QA items"):
        # Create fresh agent for each QA to avoid state issues
        agent = AssistantAgent(
            name="egolife_qa_agent",
            model_client=model_client,
            tools=tools,
            system_message=sys_prompt_text,
        )
        
        error = await process_single_qa(qa, agent, args, args.result_dir)
        if error:
            errors.append(error)
    
    # Save error list if any
    if errors:
        error_file = os.path.join(args.log_dir, f"error_list_{args.identity}.json")
        with open(error_file, "w", encoding="utf-8") as f:
            json.dump(errors, f, indent=4)
        print(f"Saved {len(errors)} errors to {error_file}")
    else:
        print("No errors encountered!")
    
    # Cleanup
    print("Processing complete!")
    if cache_dir and os.path.exists(cache_dir):
        try:
            os.rmdir(cache_dir)
        except OSError:
            pass  # Directory not empty or doesn't exist
    
    await model_client.close()


if __name__ == "__main__":
    asyncio.run(main())