[WIP-paper] Update paper

docs/paper/white_paper.tex

@@ -58,7 +58,7 @@
     want to create, then cast them on any dataset they have. In addition, these
     features can be exported into a JSON file (\textit{SpellBook}) for sharing
     and reproducibility.  Geomancer has been useful to some of our production
-    use-cases such as poverty mapping, real-estate price estimation, and more.
+    use-cases such as wealth prediction, real-estate price estimation, and more.
 \end{abstract}
 
 \section{Introduction}
@@ -73,18 +73,19 @@ \section{Introduction}
 % It's difficult to do standard data preprocessing workflows
 % -- it's difficult because it's not straightforward to do it in large-scale
 Due to this nature, feature engineering for geospatial data is not that
-straightforward. It requires a significant amount of compute and storage.
-Preliminary considerations include: (1) storage of geospatial data source, (2)
-compute necessary to query from that source, and the (3) ability to query from
-multiple data sources. If we include research requirements of
-reproducibility\cite{goodman2016does},
-then geospatial feature engineering poses to be a large-scale ML systems
-problem.
+straightforward\textemdash it requires a significant amount of compute and
+storage.  Anyone planning to perform the said task should consider (1) the
+storage required to house their geospatial data sources, (2) the compute
+necessary to query from that source, and the (3) ability to query from multiple
+data sources. If we include research requirements of
+reproducibility\cite{goodman2016does}, then geospatial feature engineering
+poses to be a large-scale ML systems problem.
 
 Geomancer\footnote{\url{https://github.com/thinkingmachines/geomancer}} offers
-a framework to perform feature engineering for geospatial data. It leverages a
-data warehouse, large-scale geospatial datasets, and off-the-shelf feature
-transforms that can be composed into more complex queries. It is open-source
+a framework to perform feature engineering for geospatial data at a large
+scale. It leverages a data warehouse, geospatial datasets, and off-the-shelf
+feature transforms that can be composed into more complex queries. It also
+provides a way to share feature transforms to other users. It is open-source
 and licensed under MIT.
 
 \subsection{On features and geospatial attributes}