CodePDE / refine.py
LDA1020's picture
feat: code release
56c4b9b verified
import os
import pandas as pd
import random
import time
from code_generation import generate_and_debug, prepare_working_folder
def select_seed_implementations(
total_num_sample_solvers,
num_sample_for_refine=None,
):
if (
num_sample_for_refine is None or
num_sample_for_refine > total_num_sample_solvers or
num_sample_for_refine == -1
):
num_sample_for_refine = total_num_sample_solvers
# Select random samples for refinement
selected_indices = random.sample(range(total_num_sample_solvers), num_sample_for_refine)
return selected_indices
def refine(cfg):
num_repeated_samples = cfg.method.num_repeated_samples
num_trials = cfg.method.num_debugging_trials_per_sample
pde_name = cfg.pde.name
working_folder = cfg.working_folder
model_name = cfg.model.name
num_sample_for_refine = cfg.method.num_sample_for_refine
start_round = cfg.method.start_round
use_sample_solver_init = cfg.method.use_sample_solver_init
assert use_sample_solver_init, 'Sample solvers must be enabled for refinement'
sample_solver_folder = os.path.join(
'solvers', pde_name, cfg.pde.pde_setting_name, 'seeds'
)
sample_solver_info = pd.read_csv(
os.path.join(sample_solver_folder, 'seed_results.csv')
)
total_num_sample_solvers = len(sample_solver_info)
if start_round == 0:
prepare_working_folder(
cfg,
working_folder=working_folder,
pde_name=pde_name,
use_sample_solver_init=use_sample_solver_init
)
for round_idx in range(start_round, num_repeated_samples):
try:
seed_implementations = select_seed_implementations(
total_num_sample_solvers=total_num_sample_solvers,
num_sample_for_refine=num_sample_for_refine
)
generate_and_debug(
cfg,
round_idx=round_idx,
num_trials=num_trials,
pde_name=pde_name,
working_folder=working_folder,
seed_implementations=seed_implementations,
model_name=model_name
)
except Exception as e:
print(f'Error in sample {round_idx}: {e}. Move on to the next sample.')
time.sleep(2) # Small delay to prevent API rate limit