Fix a bug in OrdinalEncoder

README.md

@@ -68,6 +68,12 @@ pip install tabensemb[torch]
 
 Please use `pip install tabensemb` instead if you already have `torch>=1.12.0` installed. Use `pip install tabensemb[test]` if you want to run unit tests. 
 
+To install from source,
+
+```shell
+pip install -e .[torch]
+```
+
 2. (Optional) Run unit tests after installed `tabensemb[test]`:
 
 ```shell

docs/source/api/utility.rst

@@ -11,5 +11,4 @@ tabensemb.utils
     :toctree: generated/
 
     utils
-    collate.fix_collate_fn
     ranking

requirements.txt

@@ -1,24 +1,29 @@
-autogluon.tabular[all]>=0.7.0,<1.0.0
-autogluon.common>=0.7.0,<1.0.0
-autogluon.core>=0.7.0,<1.0.0
-autogluon.features>=0.7.0,<1.0.0
+autogluon.tabular[all]>=0.8.2,<1.0.0
+autogluon.common>=0.8.2,<1.0.0
+autogluon.core>=0.8.2,<1.0.0
+autogluon.features>=0.8.2,<1.0.0
 captum>=0.6.0
-matplotlib>=3.3.4
-numpy>=1.22.1
-openpyxl>=3.0.10
-pandas>=1.4.2
-scikit-learn>=1.0.2,<1.4
+gensim>=4.3.0,<4.3.3
+matplotlib>=3.8.2
+numba>=0.58.0,<0.58.2
+numpy>=1.26.2
+opencv-contrib-python>=4.8.0,<4.8.2
+openpyxl>=3.1.2
+pandas>=1.5.3
+scikit-learn>=1.2.2,<1.4
 scikit-optimize>=0.9.0
-scipy>=1.7.3
-seaborn>=0.11.2
-tqdm>=4.64.1
-torchmetrics>=0.11.0
-pytorch-widedeep>=1.2.1
-pytorch_tabnet>=4.0
-pytorch_tabular>=1.0.2
-miceforest>=5.6.3
-shap>=0.41.0
-einops>=0.6.0
-pytorch_lightning>=1.8.6
-traitlets<=5.9.0
+scipy>=1.11.4
+seaborn>=0.13.0
+statsmodels>=0.14.0,<0.14.1
+tqdm>=4.66.1
+torchmetrics>=0.11.4
+pytorch-widedeep>=1.3.2,<1.6.2
+pytorch-tabnet>=4.0
+pytorch-tabular>=1.0.2
+ray>=2.3.1
+miceforest>=5.7.0
+shap>=0.43.0
+einops>=0.6.1
+pytorch-lightning>=1.9.5
+traitlets>=5.9.0
 

tabensemb/data/utils.py

@@ -318,7 +318,9 @@ def _inverse_transform(self, df: pd.DataFrame):
                     f"inverse-transformed."
                 )
 
-            transformed_values = np.ones_like(values).astype(self.dtypes[feature])
+            transformed_values = np.ones_like(values).astype(
+                self.dtypes[feature] if self.dtypes[feature] != str else "U256"
+            )
             for i in range(len(transformed_values)):
                 transformed_values[i] = unknown_val
             for val in unique_values:
@@ -327,5 +329,7 @@ def _inverse_transform(self, df: pd.DataFrame):
                     if val in unknown_values
                     else self.mapping[feature][int(val)]
                 )
-            df[feature] = transformed_values.astype(self.dtypes[feature])
+            df[feature] = transformed_values.astype(
+                self.dtypes[feature] if self.dtypes[feature] != str else "U256"
+            )
         return df