diff --git a/tsdart/loss.py b/tsdart/loss.py
index 498a3d2..050a52f 100644
--- a/tsdart/loss.py
+++ b/tsdart/loss.py
@@ -31,6 +31,18 @@ def __init__(self, epsilon=1e-6, mode='regularize', symmetrized=False):
         self._symmetrized = symmetrized
 
     def forward(self, data):
+        """ Compute VAMP2 loss.
+
+        Parameters
+        ----------
+        data : tuple
+            Softmax probabilities of batch of transition pairs.
+
+        Returns
+        -------
+        VAMP2 loss
+        """
+
         assert len(data) == 2
 
         koopman = estimate_koopman_matrix(data[0], data[1], epsilon=self._epsilon, mode=self._mode, symmetrized=self._symmetrized)
@@ -86,6 +98,21 @@ def __init__(self, feat_dim, n_states, device, proto_update_factor=0.5, scaling_
         self.scaling_temperature = scaling_temperature
     
     def forward(self, features, labels):
+        """ Compute dispersion loss.
+
+        Parameters
+        ----------
+        features : torch.Tensor
+            Hyperspherical embeddings of a batch of data.
+
+        labels : torch.Tensor
+            Metastable states of a batch of data. 
+
+        Returns
+        -------
+        loss : torch.Tensor
+            Dispersion loss
+        """
 
         prototypes = self.prototypes.to(device=self.device)
         for i in range(len(labels)):
@@ -143,6 +170,22 @@ def __init__(self, n_states, device, scaling_temperature=0.1):
         self.scaling_temperature = scaling_temperature
 
     def forward(self, features, labels):
+        """ Compute dispersion loss.
+
+        Parameters
+        ----------
+        features : torch.Tensor
+            Hyperspherical embeddings of a batch of data.
+
+        labels : torch.Tensor
+            Metastable states of a batch of data. 
+
+        Returns
+        -------
+        prototypes : torch.Tensor
+            State center vectors of shape [n_states, feat_dim].
+        """
+
         with torch.no_grad():
             proxy_labels = torch.arange(0, self.n_states).to(device=self.device)
             labels = labels.contiguous().view(-1, 1)