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

节点 n17

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 fbd55642fc: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 +11 −0、solution/run.py +72 −14

diff --git a/solution/METHOD.md b/solution/METHOD.mdnew file mode 100644index 0000000..c109b50--- /dev/null+++ b/solution/METHOD.md@@ -0,0 +1,11 @@+## 改了什么+相对父节点（seed pseudobulk_shift）的改动：+1. 用 `inputs_by_time(manifest, include_external=False)` 过滤外部输入阶段，修复 proxy2 上把 Qiu E9.0 当作最新官方阶段导致分数崩溃的问题（proxy2 从 27.43 恢复到 ~50）。+2. 移除无收缩的全量 delta（alpha=1），改用 SNR 自适应收缩：每个类型根据显著基因比例决定整体 alpha（0.1–0.75），每个基因根据信噪比决定收缩力度（(snr/(snr+K))^power），噪声大的基因几乎不动。+3. 单输入阶段退化为 copy_last，不再崩溃。+4. 输出前统一 clip 到 ≥0。+理由：方法卡 k018 明确指出 alpha=1 的常数位移在 T1 上低于 copy_last；实验表中所有高分节点都使用 include_external=False + SNR 收缩。+## 用到的知识与出处+- 方法卡 k018（Damped per-type shift: shrinkage alpha on the observed delta）+- 方法卡 k004（OT recipe census，确认类型配对逻辑）+- 实验表中 Program 1（score 49.08）验证了 SNR 自适应收缩 + GENE_SHRINK_POWER=2.5, K=3.0 的效果；本节点微调为 power=2.0, K=2.5 以探索不同收缩强度diff --git a/solution/run.py b/solution/run.pyindex f3a0f25..966174d 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.+"""Per-type adaptive shrinkage on pseudobulk deltas 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.+Uses only official (non-external) input stages. For each cell type, computes+the pseudobulk delta and applies SNR-based per-gene shrinkage: genes with+large signal relative to sampling noise are shifted more, noisy genes are+shrunk toward zero. Per-type alpha scales with the fraction of significant+genes. Falls back to copy_last with a single official input stage. """  from __future__ import annotations@@ -15,8 +13,8 @@ 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 (     inputs_by_time,     labels_of,@@ -28,6 +26,18 @@ from src.task1_temporal.view_io import (     write_prediction, ) +ALPHA_MAX = 0.75+ALPHA_MIN = 0.1+SIGN_THRESHOLD = 0.05+GENE_SHRINK_POWER = 2.0+GENE_SHRINK_K = 2.5+++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 +48,64 @@ 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]+    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)+        labels_last = labels_of(last)+        labels_prev = labels_of(prev)++        Xd_last = _to_dense(last.X)+        Xd_prev = _to_dense(prev.X)++        types_prev_set = set(np.unique(labels_prev))+        labels_sampled = labels_last[rows]++        for t in np.unique(labels_sampled):+            mask_s = labels_sampled == t+            if t not in types_prev_set:+                continue+            mask_last = labels_last == t+            mask_prev = labels_prev == t+            n_last = int(mask_last.sum())+            n_prev = int(mask_prev.sum())++            mean_last = Xd_last[mask_last].mean(axis=0)+            mean_prev = Xd_prev[mask_prev].mean(axis=0)+            raw_delta = mean_last - mean_prev++            var_last = Xd_last[mask_last].var(axis=0)+            var_prev = Xd_prev[mask_prev].var(axis=0)+            noise_se = np.sqrt(+                var_last / max(n_last, 1) + var_prev / max(n_prev, 1) + 1e-12+            )+            snr = np.abs(raw_delta) / (noise_se + 1e-12)+            significant = snr > 2.0+            frac_significant = significant.mean()++            if frac_significant < SIGN_THRESHOLD:+                alpha = ALPHA_MIN+                gene_shrink = np.ones_like(raw_delta)+            else:+                alpha = ALPHA_MIN + (ALPHA_MAX - ALPHA_MIN) * min(+                    frac_significant / 0.3, 1.0+                )+                gene_shrink = (snr / (snr + GENE_SHRINK_K)) ** GENE_SHRINK_POWER++            delta = (raw_delta * alpha * gene_shrink).astype(np.float32)+            X[mask_s] += delta[np.newaxis, :]++        del prev, Xd_last, Xd_prev++    np.clip(X, 0.0, None, out=X)     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

Looking at the results, time_scale=2.0 was too aggressive (X3 dropped from 46.67 to 45.71, covariation from 47.50 to 46.07). I'll use a damped time extrapolation factor and slightly increase ALPHA_MAX to compensate for the reduced extrapolation.

<<<<<<< SEARCH
ALPHA_MAX = 0.75
ALPHA_MIN = 0.1
SIGN_THRESHOLD = 0.05
GENE_SHRINK_POWER = 2.0
GENE_SHRINK_K = 2.5
=======
ALPHA_MAX = 0.80
ALPHA_MIN = 0.1
SIGN_THRESHOLD = 0.05
GENE_SHRINK_POWER = 2.0
GENE_SHRINK_K = 2.5
TIME_DAMP = 0.4
TIME_SCALE_CAP = 1.6
>>>>>>> REPLACE

<<<<<<< SEARCH
    time_scale = 1.0
    if len(stages) >= 2:
        t_target = manifest["target"]["time"]
        t_last = stages[-1]["time"]
        t_prev = stages[-2]["time"]
        dt_obs = t_last - t_prev
        if dt_obs > 1e-6:
            time_scale = min((t_target - t_last) / dt_obs, 2.0)
=======
    time_scale = 1.0
    if len(stages) >= 2:
        t_target = manifest["target"]["time"]
        t_last = stages[-1]["time"]
        t_prev = stages[-2]["time"]
        dt_obs = t_last - t_prev
        if dt_obs > 1e-6:
            raw_scale = (t_target - t_last) / dt_obs
            time_scale = min(1.0 + TIME_DAMP * (raw_scale - 1.0), TIME_SCALE_CAP)
            time_scale = max(time_scale, 0.5)
>>>>>>> REPLACE