.. docs/source/examples.rst

Examples
========

Below are examples demonstrating TimeKAN’s usage.

Simple KAN-LSTM Example
-----------------------

This script shows how to use `tKANLSTM` for time series prediction:

.. code-block:: python

   import torch
   import torch.nn as nn
   from timekan.models.tkan_lstm import tKANLSTM
   from timekan.utils.datasets import mackey_glass

   from cuda_selector import auto_cuda # Optional, must install cuda-selector


   device = auto_cuda() # Optional, cpu for no cuda or select cuda device

   class RecurrentKAN(nn.Module):
      def __init__(self, input_dim, hidden_dim):
         super().__init__()
         self.tkan = tKANLSTM(
               input_dim=input_dim,
               hidden_dim=hidden_dim,
               return_sequences=False,
               bidirectional=True,
               kan_type='fourier',
               sub_kan_configs={'gridsize': 50, 'addbias': True}
         )
         self.regressor = nn.Linear(hidden_dim * 2, 1)

      def forward(self, x):
         features = self.tkan(x)
         return self.regressor(features).squeeze(-1)

   x_train, y_train, x_test, y_test = mackey_glass()

   model = RecurrentKAN(input_dim=1, hidden_dim=16).to(device)
   criterion = nn.MSELoss()
   optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

   for epoch in range(10):
      model.train()
      optimizer.zero_grad()
      x_train, y_train = x_train.to(device), y_train.to(device)
      outputs = model(x_train)
      loss = criterion(outputs, y_train)
      loss.backward()
      optimizer.step()
      print(f"Epoch {epoch + 1}/10, Training MSE: {loss.item():.4f}")

   model.eval()
   with torch.no_grad():
      x_test, y_test = x_test.to(device), y_test.to(device)
      test_outputs = model(x_test)
      test_mse = criterion(test_outputs, y_test).item()
      print(f"Test MSE: {test_mse:.4f}")
More Examples
-------------

Check the `examples/` directory in the repository for detailed tutorials: