Digital Drosophila Archive
2.0 KiB
2.0 KiB
Phase 4T / P0 — 已封存事件时序诊断
共审计 20 条 KC 事件序列，零 credit 3 条。没有新脑模拟、权重更新或候选选择。
| split | odor | seed | KC active | target active | first target ms | peak bin ms | reason |
|---|---|---|---|---|---|---|---|
| cal | CAL_A | 54101 | 163 | 68 | 18.0 | 20 | credit_present |
| cal | CAL_A | 54102 | 154 | 62 | 16.5 | 15 | credit_present |
| cal | CAL_A | 54103 | 154 | 62 | 14.700000000000001 | 15 | credit_present |
| cal | CAL_A | 54104 | 170 | 68 | 14.3 | 15 | credit_present |
| cal | CAL_A | 54105 | 156 | 64 | 12.700000000000001 | 15 | credit_present |
| cal | CAL_B | 54101 | 165 | 55 | 14.700000000000001 | 15 | credit_present |
| cal | CAL_B | 54102 | 183 | 56 | 14.100000000000001 | 15 | credit_present |
| cal | CAL_B | 54103 | 161 | 59 | 16.3 | 15 | credit_present |
| cal | CAL_B | 54104 | 183 | 59 | 14.200000000000001 | 15 | credit_present |
| cal | CAL_B | 54105 | 143 | 61 | 13.9 | 15 | credit_present |
| held | CAL_A | 55101 | 167 | 71 | 16.1 | 15 | credit_present |
| held | CAL_A | 55102 | 164 | 69 | 20.3 | 20 | target_onset_after_US |
| held | CAL_A | 55103 | 171 | 61 | 20.200000000000003 | 20 | target_onset_after_US |
| held | CAL_A | 55104 | 155 | 75 | 16.5 | 20 | credit_present |
| held | CAL_A | 55105 | 157 | 75 | 20.1 | 25 | target_onset_after_US |
| held | CAL_B | 55101 | 161 | 58 | 16.3 | 15 | credit_present |
| held | CAL_B | 55102 | 185 | 82 | 14.8 | 15 | credit_present |
| held | CAL_B | 55103 | 171 | 58 | 18.0 | 20 | credit_present |
| held | CAL_B | 55104 | 164 | 54 | 13.100000000000001 | 15 | credit_present |
| held | CAL_B | 55105 | 166 | 63 | 15.8 | 20 | credit_present |
所有 matched/random credit 均精确复现 Phase 4S，并通过独立逐 KC 单点 oracle。 描述性曲线不能用于把旧 heldout 当成新的验证集。下一实验必须先冻结规则，再使用全新种子。 仅定位 gate 落空原因；不能把时序相关性推断为 APL 或某个回路的因果作用。