tccc1 commited on
Commit
45a2d58
·
verified ·
1 Parent(s): b853a95

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +48 -1
README.md CHANGED
@@ -181,7 +181,54 @@ pinned: false
181
  ---
182
 
183
  ## 任务九:交付目录对齐
184
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
185
 
186
 
187
  ---
 
181
  ---
182
 
183
  ## 任务九:交付目录对齐
184
+ `src/runs/stage2/`(共 24 个文件):
185
+
186
+ ```text
187
+ stage2/
188
+ ├─ thresholds.json # 四模型参数齐全
189
+ ├─ pred_roberta_val_calibrated.csv # 成员 2:RoBERTa 校准预测
190
+ ├─ pred_roberta_test_id_calibrated.csv
191
+ ├─ pred_roberta_test_ood_calibrated.csv
192
+ ├─ metrics_roberta_binary.csv # 9 行
193
+ ├─ metrics_roberta_three_way.csv # 3 行
194
+ ├─ pred_tfidf_val_calibrated.csv # 成员 1:TF-IDF 校准预测
195
+ ├─ pred_tfidf_test_id_calibrated.csv
196
+ ├─ pred_tfidf_test_ood_calibrated.csv
197
+ ├─ metrics_tfidf_binary.csv
198
+ ├─ metrics_tfidf_three_way.csv
199
+ ├─ pred_qwen_zeroshot_val_calibrated.csv # 成员 3:Qwen 零样本
200
+ ├─ pred_qwen_zeroshot_test_id_calibrated.csv
201
+ ├─ pred_qwen_zeroshot_test_ood_calibrated.csv
202
+ ├─ metrics_qwen_zeroshot_binary.csv
203
+ ├─ metrics_qwen_zeroshot_three_way.csv
204
+ ├─ pred_qwen_fewshot_val_calibrated.csv # 成员 3:Qwen 少样本
205
+ ├─ pred_qwen_fewshot_test_id_calibrated.csv
206
+ ├─ pred_qwen_fewshot_test_ood_calibrated.csv
207
+ ├─ metrics_qwen_fewshot_binary.csv
208
+ ├─ metrics_qwen_fewshot_three_way.csv
209
+ ├─ fp_by_domain.csv # 成员 2:错误分析三张表
210
+ ├─ fp_by_length.csv
211
+ └─ fp_cases.csv
212
+ ```
213
+
214
+ `figures/`(共 13 张):
215
+
216
+ ```text
217
+ figures/
218
+ ├─ confusion_binary_roberta.png # 2×2 混淆矩阵,每模型一张
219
+ ├─ confusion_binary_tfidf.png
220
+ ├─ confusion_binary_qwen_zeroshot.png
221
+ ├─ confusion_binary_qwen_fewshot.png
222
+ ├─ confusion_threeway_roberta.png # 2×3 混淆矩阵(含 Uncertain 列)
223
+ ├─ confusion_threeway_tfidf.png
224
+ ├─ confusion_threeway_qwen_zeroshot.png
225
+ ├─ confusion_threeway_qwen_fewshot.png
226
+ ├─ calibration_roberta.png # 校准图:raw vs calibrated + 每桶 n
227
+ ├─ calibration_tfidf.png
228
+ ├─ calibration_qwen_zeroshot.png
229
+ ├─ calibration_qwen_fewshot.png
230
+ └─ risk_coverage.png # 风险—覆盖率曲线(模型对比 + 参考点)
231
+ ```
232
 
233
 
234
  ---