总览 · ← 返回运行 20261001-233757-search-t1-abc-r0-C-native

节点 n5

改了什么

方法说明?节点程序自带的 METHOD.md：这个程序做了什么、为什么。

来自 programs.git c674e7d0ae:solution/METHOD.md

调研员的计划

代码改动?这个节点的程序和父节点程序的逐行差别：绿色是新增，红色是删除。

对比：父节点版本 34ea38b45b。改动的文件：solution/METHOD.md +9 −0、solution/run.py +44 −12

diff --git a/solution/METHOD.md b/solution/METHOD.mdnew file mode 100644index 0000000..ce72541--- /dev/null+++ b/solution/METHOD.md@@ -0,0 +1,9 @@+## 改了什么+1. 用 `include_external=False` 过滤外部阶段作为基准，修复 proxy2 上 Qiu E9.0 被当作最新输入的问题（proxy2 从 27.43 提升到 ~50）。+2. 对 type_deltas 施加经验贝叶斯逐基因收缩：每个基因的收缩因子 = delta²/(delta² + gene_var)，再乘以全局 ALPHA_BASE=0.5。高信噪比基因保留更多位移，噪声基因被压缩（参考 k018 的"empirical Bayes on the delta's variance"）。+3. 添加按基因标准差缩放的高斯噪声（NOISE_SCALE=0.05），保持群体协方差结构，改善 covariation 指标。+4. 无官方阶段时回退到全部阶段（兼容 X3 外部测试题）。+## 用到的知识与出处+- k018（方法卡）：收缩系数 α 与逐基因经验贝叶斯方差加权。+- k004（方法卡）：伪批量差值计算方式。+- CONTRACT 文档：proxy2 外部输入阶段的处理规则。diff --git a/solution/run.py b/solution/run.pyindex f3a0f25..4f47e2b 100644--- a/solution/run.py+++ b/solution/run.py@@ -1,13 +1,11 @@ #!/usr/bin/env python3-"""pseudobulk_shift: latest stage + per-cell-type pseudobulk delta of the last step.+"""Variance-weighted damped shift with official-stage filtering. -The delta is mean(last|type) - mean(prev|type) over the two latest inputs,-computed on the full stages and added once to a subsample of the latest stage-(clipped at 0). Types missing from the earlier stage are copied unchanged.--With a single input stage (T1 proxy: E8.5 only) there is no step to take a-delta from, so this falls back to copy_last with the same sampling. The proxy-therefore cannot tell this seed from copy_last; that gap is expected.+Fixes proxy2 by excluding external stages from base selection.+Applies per-gene empirical Bayes shrinkage on type deltas: genes with+high delta variance relative to expression variance are trusted more.+Adds small Gaussian noise scaled by per-gene residual variance to+preserve covariation structure. """  from __future__ import annotations@@ -15,6 +13,7 @@ from __future__ import annotations import argparse  import numpy as np+from scipy import sparse  from src.task1_temporal.baselines import shift_rows, type_deltas from src.task1_temporal.view_io import (@@ -28,6 +27,15 @@ from src.task1_temporal.view_io import (     write_prediction, ) +ALPHA_BASE = 0.5+NOISE_SCALE = 0.05+++def _to_dense(X):+    if sparse.issparse(X):+        return np.asarray(X.todense(), dtype=np.float32)+    return np.asarray(X, dtype=np.float32)+  def main() -> None:     parser = argparse.ArgumentParser()@@ -38,16 +46,40 @@ def main() -> None:      manifest = load_manifest(args.data)     genes = panel_genes(args.data, manifest)-    stages = inputs_by_time(manifest)++    stages = inputs_by_time(manifest, include_external=False)+    if not stages:+        stages = inputs_by_time(manifest)+     last = read_stage(args.data, stages[-1], genes)     rng = np.random.default_rng(args.seed)-    rows = sample_rows(last.n_obs, target_n_cells(manifest, last.n_obs), rng)-    X = last.X[rows]+    n_target = target_n_cells(manifest, last.n_obs)+    rows = sample_rows(last.n_obs, n_target, rng)+    X = _to_dense(last.X[rows])+     if len(stages) >= 2:         prev = read_stage(args.data, stages[-2], genes)         deltas = type_deltas(prev.X, labels_of(prev), last.X, labels_of(last))         del prev-        X = shift_rows(X, labels_of(last)[rows], deltas)++        last_dense = _to_dense(last.X)+        gene_var = np.var(last_dense, axis=0) + 1e-8++        shrunk = {}+        for ct, d in deltas.items():+            d = np.asarray(d, dtype=np.float32)+            delta_var = d ** 2+            shrink = delta_var / (delta_var + gene_var)+            shrunk[ct] = d * ALPHA_BASE * shrink+        del last_dense++        X = shift_rows(X, labels_of(last)[rows], shrunk)++        gene_std = np.sqrt(gene_var)+        noise = rng.normal(0, NOISE_SCALE, size=X.shape).astype(np.float32)+        X = X + noise * gene_std[None, :]++    X = np.clip(X, 0, None)     write_prediction(X, genes, args.out, seed=args.seed)  

调研来源?调研员查到并用到的知识条目和文献检索结果（只列标题和编号）。

没有记录调研来源。

分析结果?分析员写的 ANALYSIS.json：改了什么、各组分数怎么变、假设是否成立、经验和下一步建议。

对话摘要?每个角色和大模型对话的统计：轮数、工具调用、用时、token 数和最后的回答摘录；原始记录只给路径。

只给统计和最后回答的摘录；完整对话请到原始记录位置里列出的文件看。

分析员

ANALYSIS.json written

native

stream

<<<<<<< SEARCH
ALPHA_BASE = 0.7
NOISE_SCALE = 0.06
SHRINK_PRIOR = 0.3
=======
ALPHA_BASE = 0.5
SHRINK_PRIOR = 1.0
MIN_CELLS_FOR_DELTA = 10
>>>>>>> REPLACE

<<<<<<< SEARCH
    if len(stages) >= 2:
        prev = read_stage(args.data, stages[-2], genes)
        deltas = type_deltas(prev.X, labels_of(prev), last.X, labels_of(last))
        del prev

        last_dense = _to_dense(last.X)
        gene_var = np.var(last_dense, axis=0) + 1e-8

        shrunk = {}
        for ct, d in deltas.items():
            d = np.asarray(d, dtype=np.float32)
            delta_var = d ** 2
            shrink = delta_var / (delta_var + SHRINK_PRIOR * gene_var)
            shrunk[ct] = d * ALPHA_BASE * shrink
        del last_dense

        X = shift_rows(X, labels_of(last)[rows], shrunk)

        gene_std = np.sqrt(gene_var)
        noise = rng.normal(0, NOISE_SCALE, size=X.shape).astype(np.float32)
        X = X + noise * gene_std[None, :]
=======
    if len(stages) >= 2:
        prev = read_stage(args.data, stages[-2], genes)
        prev_labels = labels_of(prev)
        last_labels = labels_of(last)
        deltas = type_deltas(prev.X, prev_labels, last.X, last_labels)

        last_dense = _to_dense(last.X)