Merge pull request #535 from NervanaSystems/jennifer/doc

doc fixes, use overview for index page, fix icon

doc/source/axes.rst

@@ -54,6 +54,7 @@ The ``Axis`` object represents one dimension of a tensor, and can be created wit
 For tensors with multiple dimensions, we create an ``Axes`` passing in a list of individual ``Axis`` objects. Note that
 the ordering does *not* matter in specifying the axes, and has no bearing on the eventual data layout during execution. See Properties
 for a full description of axes properties.
+
   ::
 
     axes = ng.make_axes([H, W])
@@ -125,7 +126,7 @@ We can also use ``ng.cast_axis`` to recast the axes of an already defined tensor
 Properties
 ----------
 
-1. The order of Axes does not matter. ::
+1. The order of Axes does not matter.
 
   - Two tensors ``x`` and ``y`` are considered having the same type if
 
@@ -166,6 +167,7 @@ Properties
   A set of standard neon axes are defined for neon frontends.
 
   - Axes roles
+
   ::
 
     ar = ng.make_name_scope(name="ar")
@@ -177,6 +179,7 @@ Properties
     ar.Time = ng.make_axis_role()
 
   - Image / feature map
+
   ::
 
     ax = ng.make_name_scope(name="ax")
@@ -187,6 +190,7 @@ Properties
     ax.W = ng.make_axis(roles=[ar.Width], docstring="input image width")
 
   - Filter (convolution kernel)
+
   ::
 
     ax.R = ng.make_axis(roles=[ar.Height], docstring="filter height")
@@ -196,18 +200,21 @@ Properties
     ax.K = ng.make_axis(roles=[ar.Channelout], docstring="number of output feature maps")
 
   - Output
+
   ::
 
     ax.M = ng.make_axis(roles=[ar.Depth], docstring="output image depth")
     ax.P = ng.make_axis(roles=[ar.Height], docstring="output image height")
     ax.Q = ng.make_axis(roles=[ar.Width], docstring="output image width")
 
   - Recurrent
+
   ::
 
     ax.REC = ng.make_axis(roles=[ar.Time], recurrent=True, docstring="recurrent axis")
 
   - Target
+
   ::
 
     ax.Y = ng.make_axis(docstring="target")
@@ -238,7 +245,7 @@ Elementwise Binary Ops
   (C,) + (H, W) -> (C, H, W)
   (H, W) + (C,) -> (H, W, C)
 
-  In the following example, `z` from left and right are equivalent, although
+  In the following example, ``z`` from left and right are equivalent, although
   the axis orders are different.
 
   ::

doc/source/building_graphs.rst

@@ -116,7 +116,7 @@ We can guard the ``update_op`` with a context ``ng.Op.saved_user_deps`` to make
     with ng.Op.saved_user_deps():
         update_op = ng.assign(w, w + 1)
 
-This modification will then allow the `w_comp()` to properly print ``0, 0, 0`` for each call. Ops that are defined inside the context are not included in the dependencies of the computation unless explicitly named. To recreate the ``1, 2, 3`` behavior now that the ``update_op`` is guarded, we would have to explicitly name the ``update_op`` in the computation:
+This modification will then allow the ``w_comp()`` to properly print ``0, 0, 0`` for each call. Ops that are defined inside the context are not included in the dependencies of the computation unless explicitly named. To recreate the ``1, 2, 3`` behavior now that the ``update_op`` is guarded, we would have to explicitly name the ``update_op`` in the computation:
 
 .. code-block:: python
 

doc/source/index.rst

@@ -13,12 +13,7 @@
 .. limitations under the License.
 .. ---------------------------------------------------------------------------
 
-
-|Geon|
-================
-
-:Release: |version|
-:Date: |today|
+.. include:: overview.rst
 
 .. toctree::
    :hidden:

doc/source/ngraph_theme/layout.html

@@ -103,15 +103,16 @@
         <div class="wy-side-nav-search">
           {% block sidebartitle %}
 
-          {% if logo and theme_logo_only %}
+          {% if logo %}
             <a href="{{ pathto(master_doc) }}">
           {% else %}
-            <a href="{{ pathto(master_doc) }}" class="icon icon-home"> {{ project }}
+            <a href="{{ pathto(master_doc) }}" class="icon icon-home">
           {% endif %}
 
           {% if logo %}
             {# Not strictly valid HTML, but it's the only way to display/scale it properly, without weird scripting or heaps of work #}
             <img src="{{ pathto('_static/' + logo, 1) }}" class="logo" />
+	    {{ project }}
           {% endif %}
           </a>
 

doc/source/overview.rst

@@ -14,7 +14,10 @@
 .. ---------------------------------------------------------------------------
 
 Overview
-********
+========
+
+:Release: |version|
+:Date: |today|
 
 .. Note::
    Nervana Graph is currently a preview release and the API's and implementation are subject to change. We encourage you to contribute to the discussion and help shape the future Nervana Graph.
@@ -118,4 +121,4 @@ We are actively working towards:
 
 Join us
 -------
-With the rapid pace in the deep learning community we realize that a project like this won't succeed without community participation, which is what motivated us to put this preview release out there to get feedback and encourage people like you to come join us in defining the next wave of deep learning tooling. Feel free to make pull requests/suggestions/comments on `Github <https://github.com/NervanaSystems/ngraph>`_) or reach out to us on our `mailing list <https://groups.google.com/forum/#!forum/neon-users>`_. We are also hiring for full-time and internship positions.
+With the rapid pace in the deep learning community we realize that a project like this won't succeed without community participation, which is what motivated us to put this preview release out there to get feedback and encourage people like you to come join us in defining the next wave of deep learning tooling. Feel free to make pull requests/suggestions/comments on `Github <https://github.com/NervanaSystems/ngraph>`_) or reach out to us on our `mailing list <https://groups.google.com/forum/#!forum/neon-users>`_. We are also hiring for full-time and internship positions.

doc/source/transformer_usage.rst

@@ -55,7 +55,7 @@ Computation objects are created by the transformer and provide an interface to e
 Computation Creation
 --------------------
 
-Computations are created with the ``Transformer.computation`` method. When creating a computation, the user must specify a list of results which should be evaluated by the computation. These results should be ngraph ``Op``s. The transformer is able to traverse the graph backwards from these results to determine the entire subset of graph nodes required to evaluate these results, so it is not necessary for the user to specify the entire subset of nodes to execute. The user must also specify a list of graph nodes to be set as inputs to the computation. Typically these are placeholder tensors. Continuing from the above code example, a simple graph and computation can be created:
+Computations are created with the ``Transformer.computation`` method. When creating a computation, the user must specify a list of results which should be evaluated by the computation. These results should be ngraph ``Op`` s. The transformer is able to traverse the graph backwards from these results to determine the entire subset of graph nodes required to evaluate these results, so it is not necessary for the user to specify the entire subset of nodes to execute. The user must also specify a list of graph nodes to be set as inputs to the computation. Typically these are placeholder tensors. Continuing from the above code example, a simple graph and computation can be created:
 
 .. code-block:: python
 
@@ -76,7 +76,7 @@ After all computations are created, the ``Transformer.initialize`` method must b
 Computation Execution
 ---------------------
 
-This computation object can be executed with is ``__call__`` method by specifying the input ``c``.
+This computation object can be executed with its ``__call__`` method by specifying the input ``c``.
 
 .. code-block:: python
 
@@ -94,7 +94,7 @@ In real world cases, we often want computations that return multiple results. Fo
     example_comp2 = transformer.computation([d, e], b, c)
     result_d, result_e = example_comp2(2, 7)
 
-In addition to returning the final result as above, this example will also set result_d to the result of the d operation, which should be 8.
+In addition to returning the final result as above, this example will also set ``result_d`` to the result of the ``d`` operation, which should be 8.
 
 Transformed Graph State
 -----------------------
@@ -110,7 +110,7 @@ For convenience, an executor utility is provided in ngraph.util.utils. This exec
 
 .. code-block:: python
 
-    from ngraph.util.utils import executor 
+    from ngraph.util.utils import executor
     example_comp = executor(e, b, c)
     result_e = example_comp(2, 7)
 

examples/walk_through/Logistic_Regression_Part_1.ipynb

@@ -191,8 +191,6 @@
    },
    "outputs": [],
    "source": [
-    "ngt.make_transformer()\n",
-    "\n",
     "transformer = ngt.make_transformer()\n",
     "update_fun = transformer.computation([L, W, update], alpha, X, Y)"
    ]