Add docstrings for kimm.timm_utils.* (#51)

Author: james77777778

kimm/_src/utils/timm_utils.py

@@ -22,6 +22,14 @@ def _is_non_trainable_weights(name: str):
 
 @kimm_export(parent_path=["kimm.timm_utils"])
 def separate_torch_state_dict(state_dict: typing.OrderedDict):
+    """Separate the torch state dict into trainable and non-trainable parts.
+
+    Args:
+        state_dict: A `collections.OrderedDict`.
+
+    Returns:
+        A tuple containing the trainable and non-trainable state dicts.
+    """
     trainable_state_dict = state_dict.copy()
     non_trainable_state_dict = state_dict.copy()
     trainable_remove_keys = []
@@ -44,6 +52,15 @@ def separate_torch_state_dict(state_dict: typing.OrderedDict):
 
 @kimm_export(parent_path=["kimm.timm_utils"])
 def separate_keras_weights(keras_model: keras.Model):
+    """Separate the Keras model into trainable and non-trainable parts.
+
+    Args:
+        keras_model: A `keras.Model` instance.
+
+    Returns:
+        A tuple containing the trainable and non-trainable state lists. Each
+        list contains (`keras.Variable`, name) pairs.
+    """
     trainable_weights = []
     non_trainable_weights = []
     for layer in keras_model.layers:
@@ -75,6 +92,20 @@ def separate_keras_weights(keras_model: keras.Model):
 def assign_weights(
     keras_name: str, keras_weight: keras.Variable, torch_weight: np.ndarray
 ):
+    """Assign the torch weights to the keras weights based on the arguments.
+
+    Some basic criterion:
+    1. 4D must be a convolution weights (also check the name)
+    2. 2D must be a dense weights
+    3. 1D must be a vector weights
+    4. 0D must be a scalar weights
+
+    Args:
+        keras_name: A `str` representing the name of the target weights.
+        keras_weights: A `keras.Variable` representing the target weights.
+        torch_weights: A `numpy.ndarray` representing the original source
+            weights.
+    """
     if len(keras_weight.shape) == 4:
         if (
             "conv" in keras_name
@@ -119,6 +150,19 @@ def is_same_weights(
     torch_name: str,
     torch_weights: np.ndarray,
 ):
+    """Check whether the given keras weights and torch weigths are the same.
+
+    Args:
+        keras_name: A `str` representing the name of the target weights.
+        keras_weights: A `keras.Variable` representing the target weights.
+        torch_name: A `str` representing the name of the original source
+            weights.
+        torch_weights: A `numpy.ndarray` representing the original source
+            weights.
+
+    Returns:
+        A boolean indicating whether the two weights are the same.
+    """
     if np.sum(keras_weights.shape) != np.sum(torch_weights.shape):
         if np.sum(keras_weights.shape) == 0:  # Deal with scalar
             if np.sum(torch_weights.shape) == 1: