diff --git a/applications/torch_dataset/dataset.py b/applications/torch_dataset/dataset.py
new file mode 100644
index 0000000..dd361fa
--- /dev/null
+++ b/applications/torch_dataset/dataset.py
@@ -0,0 +1,140 @@
+import torch
+import numpy as np
+import pandas as pd
+from sklearn.model_selection import GroupShuffleSplit
+from mlguess.preprocessing import load_preprocessing
+from torch.utils.data import Dataset
+
+
+class CustomDataset(Dataset):
+    def __init__(self, conf, split='train'):
+        config = conf["data"]
+        data_path = config['data_path']
+        input_cols = config['input_cols']
+        output_cols = config['output_cols']
+        flat_seed = config['split_params']['flat_seed']
+        data_seed = config['split_params']['data_seed']
+        train_size = config['split_ratios']['train_size']
+        valid_size = config['split_ratios']['valid_size']
+
+        # Load data
+        data = pd.read_csv(data_path)
+        data["day"] = data["Time"].apply(lambda x: str(x).split(" ")[0])
+        data["year"] = data["Time"].apply(lambda x: str(x).split("-")[0])
+
+        # Split data into train and test
+        gsp = GroupShuffleSplit(n_splits=1, random_state=flat_seed, train_size=train_size)
+        splits = list(gsp.split(data, groups=data["year"]))
+        train_index, test_index = splits[0]
+        train_data, self.test_data = data.iloc[train_index].copy(), data.iloc[test_index].copy()
+
+        # Split train data into train and validation
+        gsp = GroupShuffleSplit(n_splits=1, random_state=flat_seed, train_size=valid_size)
+        splits = list(gsp.split(train_data, groups=train_data["year"]))
+        train_index, valid_index = splits[data_seed]
+        self.train_data, self.valid_data = train_data.iloc[train_index].copy(), train_data.iloc[valid_index].copy()
+
+        # Initialize scalers
+        self.x_scaler, self.y_scaler = load_preprocessing(conf, seed=conf["seed"])
+
+        # Fit scalers on training data
+        self.x_scaler.fit(train_data[input_cols])
+        self.y_scaler.fit(train_data[output_cols])
+
+        # Compute var on the total training data set
+        self.training_var = [
+            np.var(train_data[output_cols].values)#np.var(self.y_scaler.transform(train_data[output_cols]))
+            for i in range(self.train_data[output_cols].shape[-1])
+        ]
+
+        # Transform the individual splits
+        if split == 'train':
+            self.inputs = self.x_scaler.transform(self.train_data[input_cols])
+            self.targets = self.y_scaler.transform(self.train_data[output_cols])
+        elif split == 'valid':
+            self.inputs = self.x_scaler.transform(self.valid_data[input_cols])
+            self.targets = self.y_scaler.transform(self.valid_data[output_cols])
+        elif split == 'test':
+            self.inputs = self.x_scaler.transform(self.test_data[input_cols])
+            self.targets = self.y_scaler.transform(self.test_data[output_cols])
+        else:
+            raise ValueError("Invalid split value. Choose from 'train', 'valid', 'test'.")
+
+        self.inputs = torch.tensor(self.inputs, dtype=torch.float32)
+        self.targets = torch.tensor(self.targets, dtype=torch.float32)
+
+    def __len__(self):
+        return len(self.inputs)
+
+    def __getitem__(self, idx):
+        x = self.inputs[idx]
+        y = self.targets[idx]
+        return x, y
+
+
+if __name__ == "__main__":
+    # Example configuration dictionary
+    config = {
+        "data": {
+            "data_path": "/glade/p/cisl/aiml/ai2es/surfacelayer/cabauw_derived_data_20210720.csv",
+            "input_cols": [
+                'wind_speed:10_m:m_s-1',
+                'potential_temperature_skin_change:10_m:K_m-1',
+                'bulk_richardson:10_m:None',
+                'mixing_ratio_skin_change:2_m:g_kg-1_m-1'
+            ],
+            "output_cols": ['friction_velocity:surface:m_s-1'],
+            "split_params": {
+                "flat_seed": 42,
+                "data_seed": 0
+            },
+            "split_ratios": {
+                "train_size": 0.9,
+                "valid_size": 0.885
+            },
+            "scaler_x": {
+                "params": {
+                    "copy": True,
+                    "with_mean": True,
+                    "with_std": True
+                },
+                "type": "quantile"
+            },
+            "scaler_y": {
+                "params": {
+                    "copy": True,
+                    "with_mean": True,
+                    "with_std": True
+                },
+                "type": "normalize"
+            },
+            "batch_size": 9163,
+            "data_path": '/glade/p/cisl/aiml/ai2es/surfacelayer/cabauw_derived_data_20210720.csv'
+        },
+        "seed": 42
+    }
+
+    # Create dataset instances
+    train_dataset = CustomDataset(config, split='train')
+    valid_dataset = CustomDataset(config, split='valid')
+    test_dataset = CustomDataset(config, split='test')
+
+    x, y = train_dataset.__getitem__(0)
+
+    # Create data loaders
+    train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=config["data"]["batch_size"], shuffle=True)
+    valid_loader = torch.utils.data.DataLoader(valid_dataset, batch_size=config["data"]["batch_size"], shuffle=False)
+    test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=config["data"]["batch_size"], shuffle=False)
+
+    # Print dataset sizes
+    print(f'Train dataset size: {len(train_dataset)}')
+    print(f'Validation dataset size: {len(valid_dataset)}')
+    print(f'Test dataset size: {len(test_dataset)}')
+
+    # Example: Iterate through the train data loader
+    for batch_idx, (inputs, targets) in enumerate(train_loader):
+        print(f'Batch {batch_idx + 1}:')
+        print(f'Inputs: {inputs}')
+        print(f'Targets: {targets}')
+        if batch_idx == 1:  # Limit to 2 batches for demonstration
+            break