more readme + move things around

README.md

@@ -2,4 +2,24 @@
 
 [![Coverage Status](https://img.shields.io/coveralls/better/convoys/master.svg?style=flat)](https://coveralls.io/github/better/convoys?branch=master)
 
-See (work in progress) [documentation](https://better.engineering/convoys/).
+Convoys
+=======
+
+![pic](docs/images/dob-violations-combined.png)
+
+Convoys is a simple library that fits a few statistical model useful for modeling time-lagged conversions.
+There is a lot more info if you head over to the  [documentation](https://better.engineering/convoys/).
+
+Installation
+------------
+
+The easiest way right now is to install it straight from Github using Pip:
+
+```
+pip install -e git://github.com/better/convoys#egg=convoys
+```
+
+More info
+---------
+
+Convoys was built by [Erik Bernhardsson](https://github.com/erikbern) and has the MIT license.

docs/afterwords.rst

@@ -0,0 +1,14 @@
+Future work
+===========
+
+- Add support for a competing hazard model. This would make it possible to analyze a dataset where there's multiple final outcomes.
+- Add support for left-censoring. This would make it possible to analyze some great datasets such as `Washington Posts unsolved homicide dataset <https://github.com/washingtonpost/data-homicides>`_ where we don't know *when* a murder was dispositioned.
+- Add support for multi-stage processes where there's a number of "milestones" that depend on each other.
+- Add support for some kind of recurrent neural network or similar where the features can vary over time.
+
+
+License
+=======
+
+Convoys uses the MIT license.
+

docs/examples.rst

@@ -1,3 +1,6 @@
+Examples
+===================
+
 The main use case for something like Convoys is definitely for online advertising, ecommerce, or similar applications.
 However, there are several places these models can be useful, which we will illustrate with some very different types of datasets.
 
@@ -120,4 +123,5 @@ This will generate something like this:
 
 As you can see, marriage data does not quite follow a generalized Gamma distribution.
 This is not quite shocking, since many complex real world phenomenon are not always possible to model mathematically.
+One reason the model isn't perfect is there's an inherent bias in the dataset we use, where it only includes people who are still alive.
 Some resources `suggest <http://data.princeton.edu/pop509/ParametricSurvival.pdf>`_ fitting a "Coale-McNeil" model to this instead, which seems like an interesting future addition to Convoys.

docs/index.rst

@@ -1,23 +1,5 @@
-Convoys
-=======
-
-Convoys is a simple library that fits a few statistical model useful for modeling time-lagged conversion rates.
-You can check out the `Github repo <https://github.com/better/convoys>`_ for source code and more things!
-
-Installation
-------------
-
-The easiest way right now is to install it straight from Github using Pip:
-
-::
-
-    pip install -e git://github.com/better/convoys#egg=convoys
-
 .. include:: introduction.rst
 
-Examples
-===================
-
 .. include:: examples.rst
 
 
@@ -43,3 +25,5 @@ Full API documentation
 
 .. automodule:: convoys.gamma
    :members:
+
+.. include:: afterwords.rst

docs/introduction.rst

@@ -1,3 +1,19 @@
+Convoys
+=======
+
+Convoys is a simple library that fits a few statistical model useful for modeling time-lagged conversion rates.
+You can check out the `Github repo <https://github.com/better/convoys>`_ for source code and more things!
+
+Installation
+------------
+
+The easiest way right now is to install it straight from Github using Pip:
+
+::
+
+    pip install -e git://github.com/better/convoys#egg=convoys
+
+
 Motivation
 ----------