-
Notifications
You must be signed in to change notification settings - Fork 49
/
LLM_Geo_kernel.py
617 lines (510 loc) · 27.8 KB
/
LLM_Geo_kernel.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
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
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
import LLM_Geo_Constants as constants
import helper
import os
import requests
import networkx as nx
import pandas as pd
import geopandas as gpd
# from pyvis.network import Network
from openai import OpenAI
import configparser
import pickle
import time
import sys
import traceback
#load config
config = configparser.ConfigParser()
config.read('config.ini')
# use your KEY.
OpenAI_key = config.get('API_Key', 'OpenAI_key')
client = OpenAI(api_key=OpenAI_key)
class Solution():
"""
class for the solution. Carefully maintain it.
by Huan Ning, 2023-04-28
"""
def __init__(self,
task,
task_name,
save_dir,
role=constants.graph_role,
model=r"gpt-4o",
data_locations=[],
stream=True,
verbose=True,
):
self.task = task
self.solution_graph = None
self.graph_response = None
self.role = role
self.data_locations=data_locations
self.task_name = task_name
self.save_dir = save_dir
self.code_for_graph = ""
self.graph_file = os.path.join(self.save_dir, f"{self.task_name}.graphml")
self.source_nodes = None
self.sink_nodes = None
self.operations = [] # each operation is an element:
# {node_name: "", function_descption: "", function_definition:"", return_line:""
# operation_prompt:"", operation_code:""}
self.assembly_prompt = ""
self.parent_solution = None
self.model = model
self.stream = stream
self.verbose = verbose
self.assembly_LLM_response = ""
self.code_for_assembly = ""
self.graph_prompt = ""
self.data_locations_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(self.data_locations)])
graph_requirement = constants.graph_requirement.copy()
graph_requirement.append(f"Save the network into GraphML format, save it at: {self.graph_file}")
graph_requirement_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(graph_requirement)])
graph_prompt = f'Your role: {self.role} \n\n' + \
f'Your task: {constants.graph_task_prefix} \n {self.task} \n\n' + \
f'Your reply needs to meet these requirements: \n {graph_requirement_str} \n\n' + \
f'Your reply example: {constants.graph_reply_exmaple} \n\n' + \
f'Data locations (each data is a node): {self.data_locations_str} \n'
self.graph_prompt = graph_prompt
# self.direct_request_prompt = ''
self.direct_request_LLM_response = ''
self.direct_request_code = ''
self.chat_history = [{'role': 'system', 'content': role}]
def get_LLM_reply(self,
prompt,
verbose=True,
temperature=1,
stream=True,
retry_cnt=3,
sleep_sec=10,
system_role=None,
model=None,
):
if system_role is None:
system_role = self.role
if model is None:
model = self.model
# Query ChatGPT with the prompt
# if verbose:
# print("Geting LLM reply... \n")
count = 0
isSucceed = False
self.chat_history.append({'role': 'user', 'content': prompt})
while (not isSucceed) and (count < retry_cnt):
try:
count += 1
response = client.chat.completions.create(model=model,
# messages=self.chat_history, # Too many tokens to run.
messages=[
{"role": "system", "content": system_role},
{"role": "user", "content": prompt},
],
temperature=temperature,
stream=stream)
except Exception as e:
# logging.error(f"Error in get_LLM_reply(), will sleep {sleep_sec} seconds, then retry {count}/{retry_cnt}: \n", e)
print(f"Error in get_LLM_reply(), will sleep {sleep_sec} seconds, then retry {count}/{retry_cnt}: \n",
e)
time.sleep(sleep_sec)
response_chucks = []
if stream:
for chunk in response:
response_chucks.append(chunk)
content = chunk.choices[0].delta.content
if content is not None:
if verbose:
print(content, end='')
else:
content = response.choices[0].message.content
# print(content)
print('\n\n')
# print("Got LLM reply.")
response = response_chucks # good for saving
content = helper.extract_content_from_LLM_reply(response)
self.chat_history.append({'role': 'assistant', 'content': content})
return response
def get_LLM_response_for_graph(self, execuate=True):
# self.chat_history.append()
response = self.get_LLM_reply(
prompt=self.graph_prompt,
system_role=self.role,
model=self.model,
)
self.graph_response = response
try:
self.code_for_graph = helper.extract_code(response=self.graph_response, verbose=False)
except Exception as e:
self.code_for_graph = ""
print("Extract graph Python code rom LLM failed.")
if execuate:
exec(self.code_for_graph)
self.load_graph_file()
return self.graph_response
def load_graph_file(self, file=""):
G = None
if os.path.exists(file):
self.graph_file = file
G = nx.read_graphml(self.graph_file)
else:
if file == "" and os.path.exists(self.graph_file):
G = nx.read_graphml(self.graph_file)
else:
print("Do not find the given graph file:", file)
return None
self.solution_graph = G
self.source_nodes = helper.find_source_node(self.solution_graph)
self.sink_nodes = helper.find_sink_node(self.solution_graph)
return self.solution_graph
@property
def operation_node_names(self):
opera_node_names = []
assert self.solution_graph, "The Soluction class instance has no solution graph. Please generate the graph"
for node_name in self.solution_graph.nodes():
node = self.solution_graph.nodes[node_name]
if node['node_type'] == 'operation':
opera_node_names.append(node_name)
return opera_node_names
def get_ancestor_operations(self, node_name):
ancestor_operation_names = []
ancestor_node_names = nx.ancestors(self.solution_graph, node_name)
# for ancestor_node_name in ancestor_node_names:
ancestor_operation_names = [node_name for node_name in ancestor_node_names if node_name in self.operation_node_names]
ancestor_operation_nodes = []
for oper in self.operations:
oper_name = oper['node_name']
if oper_name in ancestor_operation_names:
ancestor_operation_nodes.append(oper)
return ancestor_operation_nodes
def get_descendant_operations(self, node_name):
descendant__operation_names = []
descendant_node_names = nx.descendants(self.solution_graph, node_name)
# for descendant_node_name in descendant_node_names:
descendant__operation_names = [node_name for node_name in descendant_node_names if node_name in self.operation_node_names]
# descendant_codes = '\n'.join([oper['operation_code'] for oper in descendant_node_names])
descendant_operation_nodes = []
for oper in self.operations:
oper_name = oper['node_name']
if oper_name in descendant__operation_names:
descendant_operation_nodes.append(oper)
return descendant_operation_nodes
def get_descendant_operations_definition(self, descendant_operations):
keys = ['node_name', 'description', 'function_definition', 'return_line']
operation_def_list = []
for node in descendant_operations:
operation_def = {key: node[key] for key in keys}
operation_def_list.append(str(operation_def))
defs = '\n'.join(operation_def_list)
return defs
def get_prompt_for_an_opearation(self, operation):
assert self.solution_graph, "Do not find solution graph!"
# operation_dict = function_def.copy()
node_name = operation['node_name']
# get ancestors code
ancestor_operations = self.get_ancestor_operations(node_name)
ancestor_operation_codes = '\n'.join([oper['operation_code'] for oper in ancestor_operations])
descendant_operations = self.get_descendant_operations(node_name)
descendant_defs = self.get_descendant_operations_definition(descendant_operations)
descendant_defs_str = str(descendant_defs)
pre_requirements = [
f'The function description is: {operation["description"]}',
f'The function definition is: {operation["function_definition"]}',
f'The function return line is: {operation["return_line"]}'
]
operation_requirement_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(
pre_requirements + constants.operation_requirement)])
operation_prompt = f'Your role: {constants.operation_role} \n\n' + \
f'operation_task: {constants.operation_task_prefix} {operation["description"]} \n\n' + \
f'This function is one step to solve the question/task: {self.task} \n\n' + \
f"This function is a operation node in a solution graph for the question/task, the Python code to build the graph is: \n{self.code_for_graph} \n\n" + \
f'Data locations: {self.data_locations_str} \n\n' + \
f'Your reply example: {constants.operation_reply_exmaple} \n\n' + \
f'Your reply needs to meet these requirements: \n {operation_requirement_str} \n\n' + \
f"The ancestor function code is (need to follow the generated file names and attribute names): \n {ancestor_operation_codes} \n\n" + \
f"The descendant function (if any) definitions for the question are (node_name is function name): \n {descendant_defs_str}"
operation['operation_prompt'] = operation_prompt
return operation_prompt
# self.operations.append(operation_dict)
# def get_prompts_for_operations(self): ######## Not use ###########
# assert self.solution_graph, "Do not find solution graph!"
# def_list, data_node_list = helper.generate_function_def_list(self.solution_graph)
#
#
# for idx, function_def in enumerate(def_list):
# operation_dict = function_def.copy()
#
# node_name = function_def['node_name']
#
# # get ancestors code
# ancestor_operations = self.get_ancestor_operations(node_name)
# ancestor_operation_codes = '\n'.join([oper['operation_code'] for oper in ancestor_operations])
# descendant_operations = self.get_descendant_operations(node_name)
# descendant_defs = self.get_descendant_operations_definition(descendant_operations)
#
# pre_requirements = [
# f'The function description is: {function_def["description"]}',
# f'The function definition is: {function_def["function_definition"]}',
# f'The function return line is: {function_def["return_line"]}'
# ]
#
# operation_requirement_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(
# pre_requirements + constants.operation_requirement)])
#
# operation_prompt = f'Your role: {constants.operation_role} \n' + \
# f'operation_task: {constants.operation_task_prefix} {function_def["description"]} \n' + \
# f'This function is one step to solve the question/task: {self.task} \n' + \
# f"This function is a operation node in a solution graph for the question/task, the Python code to build the graph is: \n{self.code_for_graph} \n" + \
# f'Data locations: {self.data_locations_str} \n' + \
# f'Reply example: {constants.operation_reply_exmaple} \n' + \
# f'Your reply needs to meet these requirements: \n {operation_requirement_str} \n \n' + \
# f"The ancestor function code is (need to follow the generated file names and attribute names): \n {ancestor_operation_codes}" + \
# f"The descendant function definitions for the question are (node_name is function name): \n {descendant_defs}"
#
#
# operation_dict['operation_prompt'] = operation_prompt
# self.operations.append(operation_dict)
# return self.operations
# initial the oepartion list
def initial_operations(self):
self.operations = []
operation_names = self.operation_node_names
for node_name in operation_names:
function_def_returns = helper.generate_function_def(node_name, self.solution_graph)
self.operations.append(function_def_returns)
def get_LLM_responses_for_operations(self, review=True):
# def_list, data_node_list = helper.generate_function_def_list(self.solution_graph)
self.initial_operations()
for idx, operation in enumerate(self.operations):
node_name = operation['node_name']
print(f"{idx + 1} / {len(self.operations)}, LLM is generating code for operation node: {operation['node_name']}")
prompt = self.get_prompt_for_an_opearation(operation)
response = self.get_LLM_reply(
prompt=prompt,
system_role=constants.operation_role,
model=self.model,
# model=r"gpt-4",
)
# print(response)
operation['response'] = response
try:
operation_code = helper.extract_code(response=operation['response'], verbose=False)
# print("operation_code:", operation_code)
except Exception as e:
operation_code = ""
operation['operation_code'] = operation_code
if review:
operation = self.ask_LLM_to_review_operation_code(operation)
return self.operations
def prompt_for_assembly_program(self):
all_operation_code_str = '\n'.join([operation['operation_code'] for operation in self.operations])
# operation_code = solution.operations[-1]['operation_code']
# assembly_prompt = f"" + \
assembly_requirement = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(constants.assembly_requirement)])
assembly_prompt = f"Your role: {constants.assembly_role} \n\n" + \
f"Your task is: use the given Python functions, return a complete Python program to solve the question: \n {self.task}" + \
f"Requirement: \n {assembly_requirement} \n\n" + \
f"Data location: \n {self.data_locations_str} \n" + \
f"Code: \n {all_operation_code_str}"
self.assembly_prompt = assembly_prompt
return self.assembly_prompt
def get_LLM_assembly_response(self, review=True):
self.prompt_for_assembly_program()
assembly_LLM_response = helper.get_LLM_reply(self.assembly_prompt,
system_role=constants.assembly_role,
model=self.model,
# model=r"gpt-4",
)
self.assembly_LLM_response = assembly_LLM_response
self.code_for_assembly = helper.extract_code(self.assembly_LLM_response)
try:
code_for_assembly = helper.extract_code(response=self.assembly_LLM_response, verbose=False)
except Exception as e:
code_for_assembly = ""
self.code_for_assembly = code_for_assembly
if review:
self.ask_LLM_to_review_assembly_code()
return self.assembly_LLM_response
def save_solution(self):
# , graph=True
new_name = os.path.join(self.save_dir, f"{self.task_name}.pkl")
with open(new_name, "wb") as f:
pickle.dump(self, f)
def get_solution_at_one_time(self):
pass
@property
def direct_request_prompt(self):
direct_request_requirement_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(
constants.direct_request_requirement)])
direct_request_prompt = f'Your role: {constants.direct_request_role} \n' + \
f'Your task: {constants.direct_request_task_prefix} to address the question or task: {self.task} \n' + \
f'Location for data you may need: {self.data_locations_str} \n' + \
f'Your reply needs to meet these requirements: \n {direct_request_requirement_str} \n'
return direct_request_prompt
def get_direct_request_LLM_response(self, review=True):
response = helper.get_LLM_reply(prompt=self.direct_request_prompt,
model=self.model,
stream=self.stream,
verbose=self.verbose,
)
self.direct_request_LLM_response = response
self.direct_request_code = helper.extract_code(response=response)
if review:
self.ask_LLM_to_review_direct_code()
return self.direct_request_LLM_response
def execute_complete_program(self, code: str, try_cnt: int = 10) -> str:
count = 0
while count < try_cnt:
print(f"\n\n-------------- Running code (trial # {count + 1}/{try_cnt}) --------------\n\n")
try:
count += 1
compiled_code = compile(code, 'Complete program', 'exec')
exec(compiled_code, globals()) # #pass only globals() not locals()
#!!!! all variables in code will become global variables! May cause huge issues! !!!!
print("\n\n--------------- Done ---------------\n\n")
return code
# except SyntaxError as err:
# error_class = err.__class__.__name__
# detail = err.args[0]
# line_number = err.lineno
#
except Exception as err:
# cl, exc, tb = sys.exc_info()
# print("An error occurred: ", traceback.extract_tb(tb))
if count == try_cnt:
print(f"Failed to execute and debug the code within {try_cnt} times.")
return code
debug_prompt = self.get_debug_prompt(exception=err, code=code)
print("Sending error information to LLM for debugging...")
# print("Prompt:\n", debug_prompt)
response = helper.get_LLM_reply(prompt=debug_prompt,
system_role=constants.debug_role,
model=self.model,
verbose=True,
stream=True,
retry_cnt=5,
)
code = helper.extract_code(response)
return code
def get_debug_prompt(self, exception, code):
etype, exc, tb = sys.exc_info()
exttb = traceback.extract_tb(tb) # Do not quite understand this part.
# https://stackoverflow.com/questions/39625465/how-do-i-retain-source-lines-in-tracebacks-when-running-dynamically-compiled-cod/39626362#39626362
## Fill the missing data:
exttb2 = [(fn, lnnr, funcname,
(code.splitlines()[lnnr - 1] if fn == 'Complete program'
else line))
for fn, lnnr, funcname, line in exttb]
# Print:
error_info_str = 'Traceback (most recent call last):\n'
for line in traceback.format_list(exttb2[1:]):
error_info_str += line
for line in traceback.format_exception_only(etype, exc):
error_info_str += line
print(f"Error_info_str: \n{error_info_str}")
# print(f"traceback.format_exc():\n{traceback.format_exc()}")
debug_requirement_str = '\n'.join([f"{idx + 1}. {line}" for idx, line in enumerate(constants.debug_requirement)])
debug_prompt = f"Your role: {constants.debug_role} \n" + \
f"Your task: correct the code of a program according to the error information, then return the corrected and completed program. \n\n" + \
f"Requirement: \n {debug_requirement_str} \n\n" + \
f"The given code is used for this task: {self.task} \n\n" + \
f"The data location associated with the given code: \n {self.data_locations_str} \n\n" + \
f"The error information for the code is: \n{str(error_info_str)} \n\n" + \
f"The code is: \n{code}"
return debug_prompt
def ask_LLM_to_review_operation_code(self, operation):
code = operation['operation_code']
operation_prompt = operation['operation_prompt']
review_requirement_str = '\n'.join(
[f"{idx + 1}. {line}" for idx, line in enumerate(constants.operation_review_requirement)])
review_prompt = f"Your role: {constants.operation_review_role} \n" + \
f"Your task: {constants.operation_review_task_prefix} \n\n" + \
f"Requirement: \n{review_requirement_str} \n\n" + \
f"The code is: \n----------\n{code}\n----------\n\n" + \
f"The requirements for the code is: \n----------\n{operation_prompt} \n----------\n"
# {node_name: "", function_descption: "", function_definition:"", return_line:""
# operation_prompt:"", operation_code:""}
print("LLM is reviewing the operation code... \n")
# print(f"review_prompt:\n{review_prompt}")
response = helper.get_LLM_reply(prompt=review_prompt,
system_role=constants.operation_review_role,
model=self.model,
verbose=True,
stream=True,
retry_cnt=5,
)
new_code = helper.extract_code(response)
reply_content = helper.extract_content_from_LLM_reply(response)
if (reply_content == "PASS") or (new_code == ""): # if no modification.
print("Code review passed, no revision.\n\n")
new_code = code
operation['code'] = new_code
return operation
def ask_LLM_to_review_assembly_code(self):
code = self.code_for_assembly
assembly_prompt = self.assembly_prompt
review_requirement_str = '\n'.join(
[f"{idx + 1}. {line}" for idx, line in enumerate(constants.assembly_review_requirement)])
review_prompt = f"Your role: {constants.assembly_review_role} \n" + \
f"Your task: {constants.assembly_review_task_prefix} \n\n" + \
f"Requirement: \n{review_requirement_str} \n\n" + \
f"The code is: \n----------\n{code} \n----------\n\n" + \
f"The requirements for the code is: \n----------\n{assembly_prompt} \n----------\n\n"
print("LLM is reviewing the assembly code... \n")
# print(f"review_prompt:\n{review_prompt}")
response = helper.get_LLM_reply(prompt=review_prompt,
system_role=constants.assembly_review_role,
model=self.model,
verbose=True,
stream=True,
retry_cnt=5,
)
new_code = helper.extract_code(response)
if (new_code == "PASS") or (new_code == ""): # if no modification.
print("Code review passed, no revision.\n\n")
new_code = code
self.code_for_assembly = new_code
def ask_LLM_to_review_direct_code(self):
code = self.direct_request_code
direct_prompt = self.direct_request_prompt
review_requirement_str = '\n'.join(
[f"{idx + 1}. {line}" for idx, line in enumerate(constants.direct_review_requirement)])
review_prompt = f"Your role: {constants.direct_review_role} \n" + \
f"Your task: {constants.direct_review_task_prefix} \n\n" + \
f"Requirement: \n{review_requirement_str} \n\n" + \
f"The code is: \n----------\n{code} \n----------\n\n" + \
f"The requirements for the code is: \n----------\n{direct_prompt} \n----------\n\n"
print("LLM is reviewing the direct request code... \n")
# print(f"review_prompt:\n{review_prompt}")
response = helper.get_LLM_reply(prompt=review_prompt,
system_role=constants.direct_review_role,
model=self.model,
verbose=True,
stream=True,
retry_cnt=5,
)
new_code = helper.extract_code(response)
if (new_code == "PASS") or (new_code == ""): # if no modification.
print("Code review passed, no revision.\n\n")
new_code = code
self.direct_request_code = new_code
def ask_LLM_to_sample_data(self, operation_code):
sampling_data_requirement_str = '\n'.join(
[f"{idx + 1}. {line}" for idx, line in enumerate(constants.sampling_data_requirement)])
sampling_data_review_prompt = f"Your role: {constants.sampling_data_role} \n" + \
f"Your task: {constants.sampling_task_prefix} \n\n" + \
f"Requirement: \n{sampling_data_requirement_str} \n\n" + \
f"The function code is: \n----------\n{code} \n----------\n\n" #+ \
# f"The requirements for the code is: \n----------\n{sampling_data_requirement_str} \n----------\n\n"
print("LLM is reviewing the direct request code... \n")
# print(f"review_prompt:\n{review_prompt}")
response = helper.get_LLM_reply(prompt=sampling_data_review_prompt,
system_role=constants.sampling_data_role,
model=self.model,
verbose=True,
stream=True,
retry_cnt=5,
)
code = helper.extract_code(response)
return code
# if (new_code == "PASS") or (new_code == ""): # if no modification.
# print("Code review passed, no revision.\n\n")
# new_code = code
# self.direct_request_code = new_code