Rows's picture
Add compact DLP baseline up to 16381
cdfe084 verified
Raw History Blame Contribute Delete
2.05 kB
from __future__ import annotations
import json
import struct
from pathlib import Path
from modchallenge.interface.base_model import ModularMultiplicationModel
class DlpTableBaseline(ModularMultiplicationModel):
def load(self, model_dir: str) -> None:
model_path = Path(model_dir)
with (model_path / "weights.json").open("r", encoding="utf-8") as file:
metadata = json.load(file)
self._prime_to_table = {}
for item in metadata["tables"]:
prime = int(item["prime"])
self._prime_to_table[str(prime)] = (
prime,
int(item["log_offset"]),
int(item["exp_offset"]),
)
self._table_bytes = (model_path / "weights.bin").read_bytes()
def preprocess_a(self, a: str) -> int:
return self._encode_supported_residue(a)
def preprocess_b(self, b: str) -> int:
return self._encode_supported_residue(b)
def preprocess_p(self, p: str) -> str:
normalized = p.strip().lstrip("0")
return normalized or "0"
def predict_digits(self, a_enc: int, b_enc: int, p_enc: str) -> list[int]:
table = self._prime_to_table.get(p_enc)
if table is None:
return [0]
prime, log_offset, exp_offset = table
a_residue = a_enc % prime
b_residue = b_enc % prime
if a_residue == 0 or b_residue == 0:
return [0]
log_a = self._read_u16(log_offset + a_residue)
log_b = self._read_u16(log_offset + b_residue)
answer_log = (log_a + log_b) % (prime - 1)
answer = self._read_u16(exp_offset + answer_log)
return [int(answer)]
def _encode_supported_residue(self, value: str) -> int:
residue = 0
for char in value.strip():
if "0" <= char <= "9":
residue = residue * 10 + ord(char) - ord("0")
return residue
def _read_u16(self, entry_offset: int) -> int:
return struct.unpack_from("<H", self._table_bytes, entry_offset * 2)[0]