Merge branch 'dev' of git.nju.edu.cn:learnware/learnware-market into dev

examples/example_image/main.py

@@ -167,11 +167,13 @@ def test_search(gamma=0.1, load_market=True):
             acc_list.append(acc)
             logger.info("search rank: %d, score: %.3f, learnware_id: %s, acc: %.3f" % (idx, score, learnware.id, acc))
         # test reuse
+        """
         reuse_baseline = JobSelectorReuser(learnware_list=mixture_learnware_list)
         reuse_predict = reuse_baseline.predict(user_data=user_data)
         reuse_score = eval_prediction(reuse_predict, user_label)
         job_selector_score_list.append(reuse_score)
         print(f"mixture reuse loss: {reuse_score}\n")
+        """
 
         reuse_ensemble = AveragingReuser(learnware_list=mixture_learnware_list, mode="vote")
         ensemble_predict_y = reuse_ensemble.predict(user_data=user_data)
@@ -186,10 +188,10 @@ def test_search(gamma=0.1, load_market=True):
         % (np.mean(select_list), np.std(select_list), np.mean(avg_list), np.std(avg_list))
     )
     logger.info("Average performance improvement: %.3f" % (np.mean(improve_list)))
-    logger.info(
-        "Average Job Selector Reuse Performance: %.3f +/- %.3f"
-        % (np.mean(job_selector_score_list), np.std(job_selector_score_list))
-    )
+    # logger.info(
+    #     "Average Job Selector Reuse Performance: %.3f +/- %.3f"
+    #     % (np.mean(job_selector_score_list), np.std(job_selector_score_list))
+    # )
     logger.info(
         "Ensemble Reuse Performance: %.3f +/- %.3f" % (np.mean(ensemble_score_list), np.std(ensemble_score_list))
     )

learnware/learnware/reuse.py

@@ -17,7 +17,7 @@ logger = get_module_logger("Reuser")
 class JobSelectorReuser(BaseReuser):
     """Baseline Multiple Learnware Reuser uing Job Selector Method"""
 
-    def __init__(self, learnware_list: List[Learnware], herding_num: int = 1000):
+    def __init__(self, learnware_list: List[Learnware], herding_num: int = 1000, use_herding: bool = True):
         """The initialization method for job selector reuser
 
         Parameters
@@ -29,6 +29,7 @@ class JobSelectorReuser(BaseReuser):
         """
         super(JobSelectorReuser, self).__init__(learnware_list)
         self.herding_num = herding_num
+        self.use_herding = use_herding
 
     def predict(self, user_data: np.ndarray) -> np.ndarray:
         """Give prediction for user data using baseline job-selector method
@@ -53,13 +54,15 @@ class JobSelectorReuser(BaseReuser):
 
         return selector_pred_y
 
-    def job_selector(self, user_data: np.ndarray):
+    def job_selector(self, user_data: np.ndarray, use_herding: bool = True):
         """Train job selector based on user's data, which predicts which learnware in the pool should be selected
 
         Parameters
         ----------
         user_data : np.ndarray
             User's labeled raw data.
+        use_herding: bool
+            Whether create job selector training samples by herding
         """
         if len(self.learnware_list) == 1:
             user_data_num = user_data.shape[0]
@@ -69,11 +72,12 @@ class JobSelectorReuser(BaseReuser):
                 learnware.specification.get_stat_spec_by_name("RKMEStatSpecification")
                 for learnware in self.learnware_list
             ]
-            task_matrix = np.zeros((len(learnware_rkme_spec_list), len(learnware_rkme_spec_list)))
 
+            task_matrix = np.zeros((len(learnware_rkme_spec_list), len(learnware_rkme_spec_list)))
             for i in range(len(self.learnware_list)):
                 task_rkme1 = learnware_rkme_spec_list[i]
-                for j in range(i, len(self.learnware_list)):
+                task_matrix[i][i] = task_rkme1.inner_prod(task_rkme1)
+                for j in range(i + 1, len(self.learnware_list)):
                     task_rkme2 = learnware_rkme_spec_list[j]
                     task_matrix[i][j] = task_matrix[j][i] = task_rkme1.inner_prod(task_rkme2)
 
@@ -87,8 +91,12 @@ class JobSelectorReuser(BaseReuser):
                 task_spec = learnware_rkme_spec_list[i]
                 task_herding_num = max(5, int(self.herding_num * task_mixture_weight[i]))
                 task_val_num = task_herding_num // 5
+                
+                if self.use_herding:
+                    herding_X_i = task_spec.herding(task_herding_num).detach().cpu().numpy()
+                else:
+                    herding_X_i = task_spec.z.detach().cpu().numpy()
 
-                herding_X_i = task_spec.herding(task_herding_num).detach().cpu().numpy()
                 train_X_i = herding_X_i[:-task_val_num]
                 val_X_i = herding_X_i[-task_val_num:]
 

learnware/market/easy.py

@@ -639,7 +639,6 @@ class EasyMarket(BaseMarket):
                 weight_list = []
                 mixture_learnware_list = []
 
-            # convert dist to score
             if mixture_dist is None:
                 sorted_score_list = self._convert_dist_to_score(sorted_dist_list)
                 mixture_score = None