Repository navigation
Expand file tree
/
Copy pathsynthesizer.py
More file actions
342 lines (279 loc) · 14.3 KB
/
Copy pathsynthesizer.py
File metadata and controls
342 lines (279 loc) · 14.3 KB
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
import re
import time
import random
from typing import Dict, Any, Optional
from dataset import ProgramSynthesisDatapoint
from model import BedrockClient
from logger import SynthesisLogger
from report import EvaluationReport
class ProgramSynthesizer:
"""Parent class for program synthesis using different prompting methods."""
def __init__(self, target_language: str, prompting_method: str,
model_name: str = "claude-haiku", api_key: Optional[str] = None,
logger: Optional[SynthesisLogger] = None):
"""
Initialize program synthesizer.
Args:
target_language: Target programming language
prompting_method: Method to use for prompting
model_name: Bedrock model ID
api_key: API key (optional, will use environment variable)
logger: SynthesisLogger instance (optional, will create default if None)
"""
self.target_language = target_language
self.prompting_method = prompting_method
self.model_name = model_name
self.model = BedrockClient(model_name, api_key)
# Use provided logger or create default one
self.logger = logger if logger is not None else SynthesisLogger()
# Store configuration
self.config = {
"target_language": target_language,
"prompting_method": prompting_method,
"model_name": model_name
}
def synthesize(self, datapoint: ProgramSynthesisDatapoint) -> str:
"""
Synthesize a program for the given datapoint using the configured prompting method.
Args:
datapoint: The problem datapoint to synthesize a solution for
Returns:
Synthesized program as string
"""
try:
# Route to appropriate prompting method
if self.prompting_method == "zero_shot":
synthesized_program = self.zero_shot(datapoint)
elif self.prompting_method == "two_step_chain_of_thought":
synthesized_program = self.two_step_chain_of_thought(datapoint)
elif self.prompting_method == "iterative_refinement":
synthesized_program = self.iterative_refinement_with_feedback(datapoint)
elif self.prompting_method == "YOUR_CUSTOM_PROMPTING_METHOD":
################################################################################
# #
# TODO: Part 2d. Implement your custom prompting method. #
# #
################################################################################
synthesized_program = self.YOUR_CUSTOM_PROMPTING_METHOD(datapoint)
else:
raise ValueError(f"Unknown prompting method: {self.prompting_method}")
# Log successful synthesis
self.logger.log_synthesis(datapoint, self.config, synthesized_program, success=True)
return synthesized_program
except Exception as e:
# Log failed synthesis
self.logger.log_synthesis(datapoint, self.config, "", success=False, error=str(e))
raise
def zero_shot(self, datapoint: ProgramSynthesisDatapoint) -> str:
################################################################################
# #
# TODO: Part 2a. Implement your zero-shot prompting method. #
# #
################################################################################
return ""
def two_step_chain_of_thought(self, datapoint: ProgramSynthesisDatapoint) -> str:
################################################################################
# #
# TODO: Part 2b. Implement your two-step chain-of-thought prompting method. #
# #
################################################################################
return ""
def iterative_refinement_with_feedback(self, datapoint: ProgramSynthesisDatapoint) -> str:
################################################################################
# #
# TODO: Part 2c. Implement your iterative refinement prompting method with #
# evaluation feedback. #
# #
################################################################################
return ""
def YOUR_CUSTOM_PROMPTING_METHOD(self, datapoint: ProgramSynthesisDatapoint) -> str:
################################################################################
# #
# TODO: Part 2d. Implement your custom prompting method. #
# #
# Note: Please change the method name to your custom prompting method. #
# #
################################################################################
return ""
def evaluate(self, datapoint: ProgramSynthesisDatapoint, synthesized_program: str) -> EvaluationReport:
"""Evaluate synthesized program (to be implemented by subclasses)."""
raise NotImplementedError("Evaluate method must be implemented by subclasses")
class MockSynthesizer:
"""Mock synthesizer for dry-run mode that simulates synthesis failures.
This class provides a MockSynthesizer that can be used during development
and testing to simulate the synthesis and evaluation pipeline without making
actual API calls to language models.
"""
def __init__(self, target_language: str, prompting_method: str, model_name: str):
"""
Initialize the mock synthesizer.
Args:
target_language: Target programming language (python, rust, ocaml)
prompting_method: Prompting method to use
model_name: Name of the model to use
"""
self.target_language = target_language
self.prompting_method = prompting_method
self.model_name = model_name
# Store configuration
self.config = {
"target_language": target_language,
"prompting_method": prompting_method,
"model_name": model_name
}
def synthesize(self, datapoint: ProgramSynthesisDatapoint) -> str:
"""
Simulate synthesis by generating a mock program that will fail.
Args:
datapoint: The problem datapoint to synthesize a solution for
Returns:
Mock synthesized program as string
"""
# Generate a mock program based on target language
if self.target_language == "python":
mock_program = self._generate_mock_python_program()
elif self.target_language == "rust":
mock_program = self._generate_mock_rust_program()
elif self.target_language == "ocaml":
mock_program = self._generate_mock_ocaml_program()
else:
mock_program = f"// Mock {self.target_language} program\n// This will fail compilation"
# Simulate some processing time
time.sleep(random.uniform(0.1, 0.5))
# Simulate LLM response with markdown code blocks
mock_response = f"""Here's my solution to the problem:
First, I need to analyze the problem requirements:
- The problem asks for a solution in {self.target_language}
- I need to consider the input/output specifications
- I should handle edge cases properly
Let me implement the solution:
```{self.target_language}
{mock_program}
```
This solution should work for the given test cases. The approach is to..."""
# Simulate code extraction
return self._extract_code_from_response(mock_response)
def _extract_code_from_response(self, response: str) -> str:
"""Simulate code extraction from LLM response."""
import re
# Look for markdown code blocks with language tags (```lang or ```{lang})
# Pattern matches: ```python, ```rust, ```ocaml, ```{python}, etc.
code_pattern = r'```(?:{})?(\w+)?\n(.*?)```'.format(self.target_language)
matches = re.findall(code_pattern, response, re.DOTALL)
if matches:
# Take the last (most recent) code block
extracted_code = matches[-1][1].strip() # matches[-1] is (lang, code)
return extracted_code
else:
# Try to find any code block with the target language
fallback_pattern = r'```(\w+)\n(.*?)```'
fallback_matches = re.findall(fallback_pattern, response, re.DOTALL)
# Look for code blocks that match our target language
for lang, code in fallback_matches:
if lang.lower() == self.target_language.lower():
extracted_code = code.strip()
return extracted_code
# If no code blocks found, return the original response
return response.strip()
def evaluate(self, datapoint: ProgramSynthesisDatapoint, synthesized_program: str) -> EvaluationReport:
"""
Simulate evaluation by creating a failed evaluation report.
Args:
datapoint: The problem datapoint
synthesized_program: The synthesized program to evaluate
Returns:
Mock evaluation report with simulated results
"""
# Simulate some processing time
time.sleep(random.uniform(0.2, 0.8))
# Create a mock evaluation report that simulates various failure modes
report = EvaluationReport(synthesized_program)
# Randomly choose failure type
failure_type = random.choice([
"synthesis_failure",
"syntax_error",
"runtime_error",
"timeout_error",
"partial_success",
])
if failure_type == "synthesis_failure":
report.synthesized = False
report.compiles = False
report.executes = False
report.add_error("synthesizer", "LLM running out of tokens")
elif failure_type == "syntax_error":
report.compiles = False
report.add_error("compiler", "SyntaxError: invalid syntax")
report.add_error("compiler", "IndentationError: unexpected indent")
elif failure_type == "runtime_error":
report.compiles = True
report.executes = False
report.add_error("runtime", "NameError: name 'undefined_var' is not defined")
elif failure_type == "timeout_error":
report.compiles = True
report.executes = True
# Add test results with timeout
for i, (test_input, expected_output) in enumerate(zip(datapoint.sample_inputs, datapoint.sample_outputs)):
if i == 0:
report.add_test_result(test_input, expected_output, "", "error", "Execution timed out")
else:
report.add_test_result(test_input, expected_output, "", "error", "Test skipped due to timeout")
elif failure_type == "partial_success":
report.compiles = True
report.executes = True
# Some tests pass, some fail
for i, (test_input, expected_output) in enumerate(zip(datapoint.sample_inputs, datapoint.sample_outputs)):
if i % 2 == 0:
# Pass half the tests
report.add_test_result(test_input, expected_output, expected_output, "passed")
else:
# Fail the other half
report.add_test_result(test_input, expected_output, "wrong_output", "failed")
else: # complete_failure
report.compiles = True
report.executes = True
# All tests fail
for test_input, expected_output in zip(datapoint.sample_inputs, datapoint.sample_outputs):
report.add_test_result(test_input, expected_output, "completely_wrong", "failed")
report.finalize()
return report
def _generate_mock_python_program(self) -> str:
"""Generate a mock Python program that will have issues."""
return '''# Mock Python program - this will have errors
def solve_problem():
# This function has syntax errors
x =
print(x)
# This will cause runtime errors
undefined_variable
# This might timeout
import time
time.sleep(10)
solve_problem()
'''
def _generate_mock_rust_program(self) -> str:
"""Generate a mock Rust program that will have issues."""
return '''// Mock Rust program - this will have compilation errors
fn main() {
// Syntax error - missing semicolon
let x = 42
// Undefined variable
println!("{}", undefined_var);
// This will cause runtime panic
let vec = vec![1, 2, 3];
println!("{}", vec[10]); // Index out of bounds
}
'''
def _generate_mock_ocaml_program(self) -> str:
"""Generate a mock OCaml program that will have issues."""
return '''(* Mock OCaml program - this will have errors *)
let solve_problem () =
(* Syntax error - missing semicolon *)
let x = 42
(* Undefined variable *)
print_endline undefined_var;
(* This will cause runtime error *)
let lst = [1; 2; 3] in
List.nth lst 10 (* Index out of bounds *)
let () = solve_problem ()
'''