[FIX] fix bugs for feature_embedding

learnware/market/heterogeneous/organizer/__init__.py

@@ -1,4 +1,5 @@
 import os
+import traceback
 import pandas as pd
 from collections import defaultdict
 from typing import List, Tuple, Union
@@ -125,6 +126,7 @@ class HeteroMapTableOrganizer(EasyOrganizer):
                 hetero_spec.save(save_path)
 
             except Exception as err:
+                traceback.print_exc()
                 logger.warning(f"Learnware {idx} generate HeteroMapTableSpecification failed! Due to {err}")
 
     def _get_hetero_learnware_ids(self, ids: Union[str, List[str]]) -> List[str]:

learnware/market/heterogeneous/organizer/hetero_map/__init__.py

@@ -39,7 +39,7 @@ class HeteroMap(nn.Module):
         temperature=10,
         base_temperature=10,
         activation="relu",
-        device="cuda:0",
+        device="cpu",
         **kwargs,
     ):
         """
@@ -174,7 +174,7 @@ class HeteroMap(nn.Module):
     def hetero_mapping(self, rkme_spec: RKMETableSpecification, features: dict) -> HeteroMapTableSpecification:
         hetero_spec = HeteroMapTableSpecification()
         data = rkme_spec.get_z()
-        cols = [features.get(str(i), "") for i in range(data.shape[1])]
+        cols = [features.get(str(i), "Unknown Feature") for i in range(data.shape[1])]
         hetero_input_df = pd.DataFrame(data=data, columns=cols)
         hetero_embedding = self._extract_batch_features(hetero_input_df)
         hetero_spec.generate_stat_spec_from_system(hetero_embedding, rkme_spec)

learnware/market/heterogeneous/organizer/hetero_map/feature_extractor.py

@@ -53,6 +53,7 @@ class NumEmbedding(nn.Module):
         x_ts : Any
             numerical features, (bs, emb_dim)
         """
+        print(np.array(col_emb).shape, np.array(x_ts).shape)
         col_emb = col_emb.unsqueeze(0).expand((x_ts.shape[0], -1, -1))
         feat_emb = col_emb * x_ts.unsqueeze(-1).float() + self.num_bias
         return feat_emb
@@ -99,13 +100,18 @@ class FeatureTokenizer:
             }
         """
         encoded_inputs = {"x_num": None, "num_col_input_ids": None}
-        num_cols = x.columns.tolist() if not shuffle else np.random.shuffle(x.columns.tolist())
-        x_num = x[num_cols].fillna(0)
 
+        num_cols = x.columns.tolist() if not shuffle else np.random.shuffle(x.columns.tolist())
+        index_cols = (
+            [i for i in range(len(x.columns))] if not shuffle else np.random.shuffle([i for i in range(len(x.columns))])
+        )
+        num_cols = [x.columns[i] for i in index_cols]
+        x_num = x.iloc(axis=1)[index_cols].fillna(0)
         if keep_input_grad:
             x_num_ts = torch.tensor(x_num.values, dtype=float, requires_grad=True)  # keep the grad
         else:
             x_num_ts = torch.tensor(x_num.values, dtype=float)
+
         num_col_ts = self.tokenizer(
             num_cols,
             padding=True,
@@ -195,9 +201,11 @@ class FeatureProcessor(nn.Module):
         **kwargs,
     ) -> Tensor:
         x_num = x_num.to(self.device)
-
+        print("?1", np.array(x_num).shape, np.array(num_col_input_ids).shape)
         num_col_emb = self.word_embedding(num_col_input_ids.to(self.device))
+        print("?2", np.array(x_num).shape, np.array(num_col_emb).shape)
         num_col_emb = self._avg_embedding_by_mask(num_col_emb, num_att_mask)
+        print("?3", np.array(x_num).shape, np.array(num_col_emb).shape)
 
         num_feat_embedding = self.num_embedding(num_col_emb, x_num)
         num_feat_embedding = self.align_layer(num_feat_embedding).float()

learnware/market/heterogeneous/searcher.py

@@ -1,5 +1,5 @@
 from typing import Tuple, List
-
+import traceback
 from ...learnware import Learnware
 from ...logger import get_module_logger
 from ..base import BaseUserInfo
@@ -34,9 +34,8 @@ class HeteroSearcher(EasySearcher):
             return True
 
         except Exception as e:
-            logger.warning(
-                f"Invalid heterogeneous search information provided. Use homogeneous search instead. Error: {e}"
-            )
+            traceback.print_exc()
+            logger.warning(f"Invalid heterogeneous search information provided. Use homogeneous search instead.")
             return False
 
     def __call__(

learnware/reuse/feature_augment.py

@@ -12,8 +12,8 @@ class FeatureAugmentReuser(BaseReuser):
     FeatureAugmentReuser is a class for augmenting features using predictions of a given learnware model and applying regression or classification on the augmented dataset.
 
     This class supports two modes:
-    - "regression": Uses RidgeCV for regression tasks.
-    - "classification": Uses LogisticRegressionCV for classification tasks.
+        - "regression": Uses RidgeCV for regression tasks.
+        - "classification": Uses LogisticRegressionCV for classification tasks.
     """
 
     def __init__(self, learnware_list: List[Learnware] = None, mode: str = None):