🔧 fix small error, update project version

pyproject.toml

@@ -9,7 +9,7 @@ packages = ["theseus"]
 
 [project]
 name = "theseus"
-version = "1.1.0"
+version = "1.2.0"
 description = "A general template for various Deep Learning tasks. Strongly relies on Pytorch"
 readme = "README.md"
 requires-python = ">=3.6"

theseus/__init__.py

@@ -11,7 +11,7 @@
 __author__ = "kaylode"
 __license__ = "MIT"
 __copyright__ = "Copyright 2020-present Kaylode"
-__version__ = "1.1.0"
+__version__ = "1.2.0"
 
 from .base import *
 from .registry import Registry

theseus/base/metrics/accuracy.py

@@ -15,15 +15,15 @@ def __init__(self, label_type: str = "multiclass", **kwargs):
         self.threshold = kwargs.get("threshold", 0.5)
         self.reset()
 
-    def update(self, output: Dict[str, Any], batch: Dict[str, Any]):
+    def update(self, outputs: Dict[str, Any], batch: Dict[str, Any]):
         """
         Perform calculation based on prediction and targets
         """
-        output = output["outputs"]
+        outputs = outputs["outputs"]
         target = batch["targets"]
 
         prediction = logits2labels(
-            output, label_type=self.type, threshold=self.threshold
+            outputs, label_type=self.type, threshold=self.threshold
         )
 
         correct = (prediction.view(-1) == target.view(-1)).sum()

theseus/tabular/base/preprocessors/splitter.py

@@ -1,5 +1,6 @@
 import os
 import os.path as osp
+import random
 
 from sklearn.model_selection import StratifiedKFold, train_test_split
 
@@ -27,6 +28,7 @@ def __init__(
             "default",
             "stratified",
             "stratifiedkfold",
+            "unique",
         ], "splitter type not supported"
 
         self.splitter_type = splitter_type
@@ -49,6 +51,10 @@ def __init__(
         elif self.splitter_type == "default":
             assert ratio is not None, "should specify ratio"
             self.ratio = ratio
+        elif self.splitter_type == "unique":
+            assert ratio is not None, "should specify ratio"
+            self.splitter = random.sample
+            self.ratio = ratio
 
     def run(self, df):
         num_samples, num_features = df.shape
@@ -63,6 +69,16 @@ def run(self, df):
             )
             train_df.to_csv(osp.join(self.save_folder, "train.csv"), index=False)
             val_df.to_csv(osp.join(self.save_folder, "val.csv"), index=False)
+        elif self.splitter_type == "unique":
+            unique_values = df[self.label_column].unique().tolist()
+            num_unique_samples = len(unique_values)
+            train_idx = self.splitter(
+                unique_values, int(num_unique_samples * self.ratio)
+            )
+            train_df = df[df[self.label_column].isin(train_idx)]
+            val_df = df[~df[self.label_column].isin(train_idx)]
+            train_df.to_csv(osp.join(self.save_folder, "train.csv"), index=False)
+            val_df.to_csv(osp.join(self.save_folder, "val.csv"), index=False)
         else:
             x, y = (
                 df.drop(self.label_column, axis=1).values,