first yadll commit

README.rst

@@ -99,8 +99,8 @@ Different networks tested on mnist:
 * Stacked Denoising Autoencoder
 * Restricted Boltzmann Machine
 * Deep Belief Network
-* RNN
-* LSTM
+* Recurent Neural Networks
+* Long Short-Term Memory
 
 get the list of available networks:
 

docs/modules/init.rst

@@ -1,4 +1,45 @@
-.. _init:
+:mod:`yadll.init`
 
-Init
-----
+.. automodule:: yadll.init
+
+Initializers
+------------
+
+.. autosummary::
+
+   constant
+   uniform
+   normal
+   glorot_uniform
+   glorot_normal
+   he_uniform
+   he_normal
+   orthogonal
+
+
+Detailed description
+--------------------
+
+.. autoclass:: initializer
+   :members:
+
+.. autoclass:: constant
+   :members:
+
+.. autoclass:: uniform
+   :members:
+
+.. autoclass:: normal
+   :members:
+
+.. autoclass:: glorot_uniform
+   :members:
+
+.. autoclass:: glorot_normal
+   :members:
+
+.. autoclass:: he_uniform
+   :members:
+
+.. autoclass:: he_normal
+   :members:

yadll/init.py

@@ -10,18 +10,55 @@
 
 
 def initializer(init_obj, shape, name, fan=None):
+    """
+    Call an Initializer from an init_obj
+
+    Parameters
+    ----------
+    init_obj : `init_obj`
+            an init_obj is an initializer function or the tuple of (initializer function, dict of args)
+            example : init_obj = glorot_uniform  or init_obj = (glorot_uniform, {'gain':tanh, 'borrow':False})
+    shape : `tuple` or int
+        shape of the return shared variables
+    Returns
+    -------
+        Initialized shared variables
+    """
     if not isinstance(init_obj, tuple):
         return init_obj(shape, name=name, fan=fan)
     else:
         return init_obj[0](shape, name=name, fan=fan, **init_obj[1])
 
 
 def constant(shape, value=0.0, name=None, borrow=True, **kwargs):
+    """
+    Initialize all the weights to a constant value
+
+    Parameters
+    ----------
+    shape
+    scale
+    """
     return shared_variable(np.ones(shape=shape) * value,
                            name=name, borrow=borrow)
 
 
 def uniform(shape, scale=0.5, name=None, borrow=True, **kwargs):
+    """
+    Initialize all the weights from the uniform distribution
+
+    Parameters
+    ----------
+    shape
+    scale
+    name
+    borrow
+    kwargs
+
+    Returns
+    -------
+
+    """
     if not isinstance(scale, tuple):
         scale = (-scale, scale)      # (low, high)
     return shared_variable(np_rng.uniform(low=scale[0], high=scale[1], size=shape),