|
import json |
|
from typing import List |
|
from tqdm import tqdm |
|
from random import randint, random, sample |
|
from astpretty import pprint |
|
from copy import deepcopy |
|
from string import ascii_letters |
|
import ast |
|
|
|
|
|
def write_rows(rows): |
|
|
|
with open('data.10_simple_variations.jsonl', 'a') as f: |
|
f.writelines(rows) |
|
|
|
|
|
class AlternativeList(ast.NodeTransformer): |
|
def __init__(self): |
|
self.n_changes = 0 |
|
super().__init__() |
|
|
|
def visit_List(self, node: List): |
|
code = ast.unparse(node) |
|
try: |
|
list_val = eval(code) |
|
except Exception: |
|
return node |
|
|
|
for _ in range(randint(0, 3)): |
|
list_val.append(randint(-999, 999)) |
|
|
|
if not list_val: |
|
list_val.append(randint(-999, 999)) |
|
|
|
list_val = sample(list_val, randint(1, len(list_val))) |
|
|
|
self.n_changes += 1 |
|
return ast.parse(str(list_val)).body[0].value |
|
|
|
|
|
class AlternativeConstant(ast.NodeTransformer): |
|
def __init__(self): |
|
self.n_changes = 0 |
|
super().__init__() |
|
|
|
def visit_Constant(self, node): |
|
''' |
|
Switch constant values to simple variations |
|
''' |
|
if type(node.value) is int: |
|
if randint(0, 1) == 1: |
|
node.value = randint(-9, 9) |
|
else: |
|
node.value = randint(-999, 999) |
|
elif type(node.value) is str: |
|
if randint(0, 1) == 1 and node.value: |
|
if node.value: |
|
node.value = ''.join(sample(node.value, randint(1, len(node.value)))) |
|
else: |
|
self.n_changes -= 1 |
|
else: |
|
node.value = ''.join(sample(ascii_letters, randint(1, 4))) |
|
elif type(node.value) is float: |
|
if randint(0, 1) == 1: |
|
node.value = random() |
|
else: |
|
node.value = random() * randint(-999, 999) |
|
elif type(node.value) is bool: |
|
node.value = bool(randint(0, 1)) |
|
else: |
|
self.n_changes -= 1 |
|
self.n_changes += 1 |
|
return super().visit_Constant(node) |
|
|
|
|
|
class AlternativeNames(ast.NodeTransformer): |
|
|
|
def visit_Name(self, node): |
|
return ast.copy_location(ast.Subscript( |
|
value=ast.Name(id='data', ctx=ast.Load()), |
|
slice=ast.Index(value=ast.Str(s=node.id)), |
|
ctx=node.ctx |
|
), node) |
|
|
|
|
|
def state_dict_to_str(state): |
|
vals = [] |
|
for k, v in state.items(): |
|
vals.append( |
|
f'{k} = {v}' |
|
) |
|
vals = sorted(vals) |
|
return ';'.join(vals) |
|
|
|
|
|
def trace_code(start_state: str, code: str): |
|
state = {} |
|
try: |
|
exec(start_state, {}, state) |
|
except Exception: |
|
return |
|
start_state = dict(state) |
|
try: |
|
exec(code, {}, state) |
|
except Exception: |
|
return |
|
return state_dict_to_str(start_state), code, state_dict_to_str(state) |
|
|
|
|
|
def make_alternative_rows(start, code): |
|
variations = {} |
|
n_tries = 0 |
|
state_root = ast.parse(start) |
|
|
|
while len(variations) < 10 and n_tries < 20: |
|
|
|
node_transformer = AlternativeList() |
|
alt_state_root = node_transformer.visit(deepcopy(state_root)) |
|
|
|
if node_transformer.n_changes < 1: |
|
node_transformer = AlternativeConstant() |
|
alt_state_root = node_transformer.visit(deepcopy(alt_state_root)) |
|
if node_transformer.n_changes < 1: |
|
n_tries += 10 |
|
|
|
alt_start = ast.unparse(alt_state_root) |
|
|
|
alt_start_code_end = trace_code(alt_start, code) |
|
if alt_start_code_end: |
|
variations[alt_start] = alt_start_code_end |
|
n_tries += 1 |
|
|
|
|
|
''' |
|
get number of vals in start states (N) |
|
get N random lowercase letters in alphabetical order |
|
parse start state and shuffle the expr.body |
|
make name map old->new using new characters |
|
use visitor with name map to swap variable names using map |
|
do this for start, code, end seperately |
|
''' |
|
return [ |
|
json.dumps({'start': st, 'code': cd, 'end': en}) + '\n' for st, cd, en in variations.values() |
|
] |
|
|
|
|
|
STATS = { |
|
'alt_count': 0, |
|
'num_rows': 0 |
|
} |
|
|
|
|
|
with open('new_all_states.txt', 'r') as f: |
|
rows = [] |
|
prev_lines = [] |
|
for i, line in tqdm(enumerate(f), total=9_000_000): |
|
line = line.strip() |
|
|
|
if line[-1] != ';': |
|
prev_lines.append(line) |
|
continue |
|
elif prev_lines: |
|
line = '\n'.join(prev_lines) + '\n' + line |
|
prev_lines = [] |
|
|
|
start, code_end = line.split('; code: ') |
|
start = start.removeprefix('state: ') |
|
code, end = code_end.split('; output: ') |
|
end = end.removesuffix(';') |
|
|
|
rows.append(json.dumps({ |
|
'start': start, 'code': code, 'end': end |
|
}) + '\n') |
|
alt_rows = make_alternative_rows(start, code) |
|
rows += alt_rows |
|
|
|
STATS['alt_count'] += len(alt_rows) |
|
STATS['num_rows'] += 1 |
|
|
|
if len(rows) > 100_000: |
|
breakpoint() |
|
write_rows(rows) |
|
rows = [] |
|
|