总览 · ← 返回运行 20261002-202908-search-t1-scr-C

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 251065a1ce: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 5c87608489。改动的文件：solution/METHOD.md +9 −0、solution/run.py +39 −3

diff --git a/solution/METHOD.md b/solution/METHOD.mdnew file mode 100644index 0000000..ed014d5--- /dev/null+++ b/solution/METHOD.md@@ -0,0 +1,9 @@+## 改了什么+相对父节点 copy_last：增加 per-type pseudobulk shift with per-gene empirical Bayes shrinkage。当视图有 ≥2 个输入阶段时，对每个在两阶段都出现的细胞类型计算伪批量均值差 delta，再按基因计算 shrinkage = delta^2 / (delta^2 + var_within + eps)，其中 var_within 是该基因在最后阶段该类型内的方差。有效位移 = alpha * shrinkage * delta，clip ≥0。高方差基因（主导协方差结构）几乎不动，清晰 DE 基因获得接近 alpha 的位移。alpha=0.5。单输入退化为 copy_last。++第 0 轮教训：均匀位移（无 per-gene shrinkage，alpha=0.35）使 covariation 从 48.44 暴跌到 27.5，因为所有基因等量平移破坏了基因间共变结构。本轮加入 per-gene shrinkage 后，高方差基因（对协方差贡献最大）的位移被压缩到接近 0，预期恢复 covariation 同时保留 de_recovery 的改善。++## 用到的知识与出处+- 方法卡 k018：per-gene empirical Bayes shrinkage（"shrink per gene so noisy genes move less"）。+- 方法卡 k012：pseudobulk_shift 定义；covariation 指标对方差塌缩敏感。+- 第 0 轮实验反馈：均匀位移导致 covariation 崩溃（48.44→27.5），de_recovery 改善（44.09→46.9）。diff --git a/solution/run.py b/solution/run.pyindex aa4bbf0..da905ca 100644--- a/solution/run.py+++ b/solution/run.py@@ -1,11 +1,12 @@ #!/usr/bin/env python3-"""copy_last: the latest input stage, subsampled into the board's cell range."""+"""Per-type shift with per-gene empirical Bayes shrinkage; copy_last fallback."""  from __future__ import annotations  import argparse  import numpy as np+from scipy import sparse  from src.task1_temporal.view_io import (     inputs_by_time,@@ -17,6 +18,33 @@ from src.task1_temporal.view_io import (     write_prediction, ) +ALPHA = 0.5+EPS = 1e-8+++def _pseudobulk_shift(last, prev, alpha: float) -> sparse.csr_matrix:+    last_types = np.asarray(last.obs["celltype"])+    prev_types = np.asarray(prev.obs["celltype"])++    X_last = last.X.toarray() if sparse.issparse(last.X) else np.asarray(last.X, dtype=np.float32)+    X_prev = prev.X.toarray() if sparse.issparse(prev.X) else np.asarray(prev.X, dtype=np.float32)++    shift = np.zeros_like(X_last)+    for ct in np.unique(last_types):+        m_last = last_types == ct+        m_prev = prev_types == ct+        if m_prev.sum() == 0:+            continue+        mean_last = X_last[m_last].mean(axis=0)+        mean_prev = X_prev[m_prev].mean(axis=0)+        delta = mean_last - mean_prev+        var_within = X_last[m_last].var(axis=0)+        shrinkage = delta ** 2 / (delta ** 2 + var_within + EPS)+        shift[m_last] = alpha * shrinkage * delta++    X_out = np.clip(X_last + shift, 0, None).astype(np.float32)+    return sparse.csr_matrix(X_out)+  def main() -> None:     parser = argparse.ArgumentParser()@@ -27,10 +55,18 @@ def main() -> None:      manifest = load_manifest(args.data)     genes = panel_genes(args.data, manifest)-    last = read_stage(args.data, inputs_by_time(manifest)[-1], genes)+    entries = inputs_by_time(manifest)+    last = read_stage(args.data, entries[-1], genes)     rng = np.random.default_rng(args.seed)++    if len(entries) >= 2:+        prev = read_stage(args.data, entries[-2], genes)+        X = _pseudobulk_shift(last, prev, ALPHA)+    else:+        X = last.X+     rows = sample_rows(last.n_obs, target_n_cells(manifest, last.n_obs), rng)-    write_prediction(last.X[rows], genes, args.out, seed=args.seed)+    write_prediction(X[rows], genes, args.out, seed=args.seed)   if __name__ == "__main__":

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

Looking at the results:
- Round 0 (uniform shift, α=0.35): covariation crashed (27.5), de_recovery improved (46.9), score 44.66
- Round 1 (per-gene EB shrinkage, α=0.5): covariation recovered (45.43 vs 48.44 baseline), but de_recovery barely improved (44.17 vs 44.09), score 47.44 vs baseline 47.92

The per-gene shrinkage is too aggressive — it's killing most of the signal. The shrinkage formula `delta²/(delta²+var_within)` is very conservative because within-cell variance in scRNA-seq is huge relative to pseudobulk differences. I need a middle ground.

Key insight: the problem is that `var_within` is computed on single-cell data (huge noise), making shrinkage ≈ 0 for almost all genes. Instead, I should use the variance of the *pseudobulk mean estimate* (var_within / n_cells), which gives a proper signal-to-noise ratio for the mean difference. This is the correct empirical Bayes formulation: shrinkage = delta² / (delta² + SE²) where SE² = var_within/n.

Also, I'll reduce alpha slightly to 0.4 to be more conservative on covariation while the shrinkage formula properly handles which genes to move.

<<<<<<< SEARCH
#!/usr/bin/env python3
"""Per-type shift with per-gene empirical Bayes shrinkage; copy_last fallback."""

from __future__ import annotations

import argparse

import numpy as np
from scipy import sparse

from src.task1_temporal.view_io import (
    inputs_by_time,
    load_manifest,
    panel_genes,
    read_stage,
    sample_rows,
    target_n_cells,
    write_prediction,
)

ALPHA = 0.5
EPS = 1e-8


def _pseudobulk_shift(last, prev, alpha: float) -> sparse.csr_matrix:
    last_types = np.asarray(last.obs["celltype"])
    prev_types = np.asarray(prev.obs["celltype"])

    X_last = last.X.toarray() if sparse.issparse(last.X) else np.asarray(last.X, dtype=np.float32)