DOC: Addresses multiple doc related issues

src/gfdl/model.py

@@ -136,7 +136,7 @@ class GFDLClassifier(ClassifierMixin, GFDL):
 
     Parameters
     ----------
-    hidden_layer_sizes : array-like of shape(n_layers,), default=(100,)
+    hidden_layer_sizes : array-like of shape (n_layers,), default=(100,)
         The ith element represents the number of neurons in the ith
         hidden layer.
 
@@ -169,7 +169,8 @@ class GFDLClassifier(ClassifierMixin, GFDL):
 
         - 'zeros': set weights to zeros (:func:`zeros <gfdl.weights.zeros>`).
 
-        - 'range': discrete uniform distribution (:func:`range <gfdl.weights.range>`).
+        - 'range': set weights to normalized np.arange
+          (:func:`range <gfdl.weights.range>`).
 
         - 'uniform': uniform distribution (:func:`uniform <gfdl.weights.uniform>`).
 
@@ -232,7 +233,7 @@ class GFDLClassifier(ClassifierMixin, GFDL):
         bias term corresponding to layer i.
 
     coeff_ : ndarray of shape (n_features_out, n_outputs)
-        Output weight matrix learned by linear regression.
+        Output weight matrix learned by fit method.
 
     See Also
     --------
@@ -480,7 +481,8 @@ class EnsembleGFDLClassifier(ClassifierMixin, EnsembleGFDL):
 
         - 'zeros': set weights to zeros (:func:`zeros <gfdl.weights.zeros>`).
 
-        - 'range': discrete uniform distribution (:func:`range <gfdl.weights.range>`).
+        - 'range': set weights to normalized np.arange
+          (:func:`range <gfdl.weights.range>`).
 
         - 'uniform': uniform distribution (:func:`uniform <gfdl.weights.uniform>`).
 
@@ -674,7 +676,7 @@ class GFDLRegressor(RegressorMixin, MultiOutputMixin, GFDL):
 
     Parameters
     ----------
-    hidden_layer_sizes : array-like of shape(n_layers,), default=(100,)
+    hidden_layer_sizes : array-like of shape (n_layers,), default=(100,)
         The ith element represents the number of neurons in the ith
         hidden layer.
 
@@ -707,7 +709,8 @@ class GFDLRegressor(RegressorMixin, MultiOutputMixin, GFDL):
 
         - 'zeros': set weights to zeros (:func:`zeros <gfdl.weights.zeros>`).
 
-        - 'range': discrete uniform distribution (:func:`range <gfdl.weights.range>`).
+        - 'range': set weights to normalized np.arange
+          (:func:`range <gfdl.weights.range>`).
 
         - 'uniform': uniform distribution (:func:`uniform <gfdl.weights.uniform>`).
 
@@ -768,7 +771,7 @@ class GFDLRegressor(RegressorMixin, MultiOutputMixin, GFDL):
         bias term corresponding to layer i.
 
     coeff_ : ndarray of shape (n_features_out, n_outputs)
-        Output weight matrix learned by linear regression.
+        Output weight matrix learned by the fit method.
 
     See Also
     --------

src/gfdl/weights.py

@@ -29,7 +29,8 @@ def zeros(d, h, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
     """
     return np.zeros((h, d))
 
@@ -58,20 +59,15 @@ def uniform(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
-
-    Notes
-    -----
-    uniform() had to be split out for pickle/serialization
-    for conformance with the sklearn estimator API:
-    https://scikit-learn.org/stable/developers/develop.html#developing-scikit-learn-estimators
+        Other keyword arguments. Placeholder for exposing distribution
+        parameters later on.
     """
     return rng.uniform(0, 1, (h, d))
 
 
 def range(d, h, **kwargs):
     """
-    The weight function returning samples drawn from discrete uniform distribution.
+    The weight function setting weights to a normalized np.arange.
 
     Parameters
     ----------
@@ -85,12 +81,13 @@ def range(d, h, **kwargs):
     Returns
     -------
     ndarray or scalar
-        Drawn samples from the discrete uniform distribution ``[0, d*h)``.
+        Set the weights to normalized np.arange over the range ``[0, d*h)``.
 
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
     """
     s = np.arange(d * h)
     s = np.subtract(s, np.mean(s))
@@ -124,20 +121,12 @@ def he_uniform(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----
-    This implementation deviates from the standard expression where
-    the number of input features (d) are used to compute the limit.
     https://faroit.com/keras-docs/2.0.0/initializers/#he_uniform
-    However, using the standard form returned a different
-    answer from GrafoRVFL, which uses the output size i.e. hidden
-    layer size instead (from ChatGPT). Needs further exploration
-    of why they deviate from the standard form.
-    If we choose to use the standard form, then our tests cannot be
-    used to compare against GrafoRVFL as the results could be order
-    one difference or higher.
     """
 
     limit = np.sqrt(6 / h)
@@ -169,11 +158,11 @@ def lecun_uniform(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----
-    Same comment as "he_uniform"
     https://faroit.com/keras-docs/2.0.0/initializers/#lecun_uniform
     """
 
@@ -206,7 +195,8 @@ def glorot_uniform(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----
@@ -243,7 +233,8 @@ def normal(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Other keyword arguments. Placeholder for exposing distribution
+        parameters later on.
     """
     return rng.normal(0, 1, (h, d))
 
@@ -268,16 +259,16 @@ def he_normal(d, h, *, rng, **kwargs):
     -------
     ndarray or scalar
         Draw samples from the He normal distribution with
-        mean ``0`` and standard deviation ``sqrt(2/d)``.
+        mean ``0`` and standard deviation ``sqrt(2/h)``.
 
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----
-    Same comment as "he_uniform"
     https://faroit.com/keras-docs/2.0.0/initializers/#he_normal
     """
 
@@ -305,16 +296,16 @@ def lecun_normal(d, h, *, rng, **kwargs):
     -------
     ndarray or scalar
         Draw samples from the Lecun normal distribution
-        with mean ``0`` and standard deviation ``1/sqrt(h)``.
+        with mean ``0`` and standard deviation ``sqrt(1/h)``.
 
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----
-    Same comment as "he_uniform"
     https://www.tensorflow.org/api_docs/python/tf/keras/initializers/LecunNormal
     """
 
@@ -347,7 +338,8 @@ def glorot_normal(d, h, *, rng, **kwargs):
     Other Parameters
     ----------------
     **kwargs : dict
-        Other keyword arguments.
+        Needed for keyword arguments and compatibility with other weight function apis
+        but not relevant for this function.
 
     Notes
     -----