File size: 1,837 Bytes
f7b6133 | 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 | from __future__ import annotations
from dataclasses import dataclass, asdict
@dataclass(frozen=True)
class FrozenConfig:
"""Frozen MS-MARCO-developed configuration.
The point of the six-dataset campaign is transfer, not per-dataset tuning.
Only corpus-mechanical settings such as max_features/min_df should be changed
when a dataset physically requires it.
"""
# Sparse lexical representation
max_features: int = 50_000
min_df: int = 1
lowercase: bool = True
token_pattern: str = r"(?u)\b\w\w+\b"
# Fuzzy index
F: int = 4 # fuzzy memberships/document
B: int = 64 # sparse center support
S: int = 16 # signed residual support/membership
# Reliability
tau: float = 20.0
beta: float = -0.2
reliability_eps: float = 1e-6
# Corpus term geometry
L: int = 12 # top terms/document used to estimate graph
graph_significance_tau: float = 10.0
assoc_k: int = 64 # first-order PPMI neighbors retained
route_k: int = 32 # second-order context neighbors retained
graph_block_size: int = 128
# Query routing
route_alpha: float = 0.10
route_budget: int = 32 # strongest total route coordinates; original terms preserved
# Head / tail scoring
head_k: int = 10
gamma_head: float = 0.5
gamma_tail: float = 1.0
lambda_membership: float = 2.0
# Final binary-support reranker
rerank_pool: int = 2_000
lambda_lex: float = 2.5
length_b: float = 0.2
semantic_k: int = 16
lambda_sem: float = 0.05
# Requested output depth
output_k: int = 100
def to_dict(self) -> dict:
return asdict(self)
@classmethod
def from_dict(cls, d: dict) -> "FrozenConfig":
return cls(**d)
|