Merge pull request #19 from facebookresearch/update-temporaries-docs

Update docs and write about temporaties explicitly

Author: prigoyal

docs/source/docker_image.rst

@@ -35,9 +35,9 @@ The conda and non-conda images for each setup are below:
 TC runtime image with nvidia-docker
 -----------------------------------
 
-We also provide a runtime nvidia-docker image for `Ubuntu 14.04`, gcc 4.8, CUDA 8 and CUDNN 6. Using this image,
-you can also run gpu tests. To run the image, make sure you have `nvidia-docker` installed. Then run the image using
-following command:
+We also provide a runtime nvidia-docker image for :code:`Ubuntu 14.04`, :code:`gcc 4.8`, :code:`CUDA 8` and :code:`CUDNN 6`.
+Using this image, you can also run gpu tests. To run the image, make sure you
+have :code:`nvidia-docker` installed. Then run the image using following command:
 
 * :code:`NVIDIA-Docker Ubuntu 14.04 conda environment`
 

docs/source/installation.rst

@@ -1,6 +1,8 @@
 Installation Guide
 ==================
 
+**Author**: `Priya Goyal <https://github.com/prigoyal>`_
+
 The following instructions are provided for developers who would like to
 experiment with the library.
 
@@ -16,4 +18,4 @@ Some users might prefer building TC in :code:`non-conda` enviroment and some mig
 
 Further, we also provide runtime :code:`docker` images for both :code:`conda` and :code:`non-conda` environment and also an :code:`nvidia-docker` runtime image for TC to have access to GPUs.
 
-You can chose whatever build settings suit your requirements best and follow the instructions to build.
+You can chose whatever build settings suit your requirements best and follow the instructions to build. Please feel free to contact us in case you need help with build.

docs/source/ml_with_tc.rst

@@ -38,6 +38,21 @@ scope in a TC, allocations etc (i.e. all the basic stuff you expect from a
 Programming Language). As a consequence TC is not a Programming Language but a concise
 notation. For now, it should not try to be a Programming Language.
 
+As a result of this, everything in TC should be either an input or output. For example:
+consider the TC definition below:
+
+.. code::
+
+    def softmax(float(N, D) I) -> (O, expsum) {
+      expsum(n) +=! exp(I(n, d))
+      O(n, d) = exp(I(n, d)) / expsum(n)
+    }
+
+In this TC, :code:`expsum` is a temporary variable that needs to be computed but
+since TC doesn't do allocations itself, we set it as another output. User can chose
+to ignore this output. We will work on enhancing this and deal with temporary
+allocations better in future.
+
 Graph Level
 ^^^^^^^^^^^
 
@@ -79,7 +94,7 @@ TC
 ^^
 
 The current TC implementation sits somewhere here; less verbose than Halide,
-more verbose than matrix algebra. The inference procedure has been one subtle 
+more verbose than matrix algebra. The inference procedure has been one subtle
 tradeoff in TC. It has been designed to follow an intuitive enough mental model,
 but may still evolve in the future towards greater expressiveness, see :ref:`inference`.