Make clusters look better by averaging locations of clustered points

man/tippecanoe.1

@@ -3,6 +3,8 @@
 Builds vector tilesets \[la]https://www.mapbox.com/developers/vector-tiles/\[ra] from large (or small) collections of GeoJSON \[la]http://geojson.org/\[ra], Geobuf \[la]https://github.com/mapbox/geobuf\[ra], or CSV \[la]https://en.wikipedia.org/wiki/Comma-separated_values\[ra] features,
 like these \[la]MADE_WITH.md\[ra]\&.
 .PP
+[Mapbox Tippecanoe](\[la]https://user-images.githubusercontent.com/1951835/36568734-ede27ec0-17df-11e8-8c22-ffaaebb8daf4.JPG\[ra])
+.PP
 [Build Status](https://travis\-ci.org/mapbox/tippecanoe.svg) \[la]https://travis-ci.org/mapbox/tippecanoe\[ra]
 [Coverage Status](https://coveralls.io/repos/mapbox/tippecanoe/badge.svg?branch=master&service=github) \[la]https://coveralls.io/github/mapbox/tippecanoe?branch=master\[ra]
 .SH Intent
@@ -212,7 +214,8 @@ If the type is \fB\fCfloat\fR or \fB\fCint\fR and the original attribute was non
 If the type is \fB\fCint\fR and the original attribute was floating\-point, it is rounded to the nearest integer.
 .IP \(bu 2
 \fB\fC\-E\fR\fIattribute\fP\fB\fC:\fR\fIoperation\fP or \fB\fC\-\-accumulate\-attribute=\fR\fIattribute\fP\fB\fC:\fR\fIoperation\fP: Preserve the named \fIattribute\fP from features
-that are dropped, coalesced\-as\-needed, or clustered. The \fIoperation\fP may be \fB\fCsum\fR, \fB\fCproduct\fR, \fB\fCmean\fR, \fB\fCconcat\fR, or \fB\fCcomma\fR
+that are dropped, coalesced\-as\-needed, or clustered. The \fIoperation\fP may be
+\fB\fCsum\fR, \fB\fCproduct\fR, \fB\fCmean\fR, \fB\fCmax\fR, \fB\fCmin\fR, \fB\fCconcat\fR, or \fB\fCcomma\fR
 to specify how the named \fIattribute\fP is accumulated onto the attribute of the same name in a feature that does survive.
 .RE
 .SS Filtering features by attributes
@@ -247,7 +250,7 @@ compensate for the larger marker, or \fB\fC\-Bf\fR\fInumber\fP to allow at most
 .IP \(bu 2
 \fB\fC\-ap\fR or \fB\fC\-\-drop\-polygons\fR: Let "dot" dropping at lower zooms apply to polygons too
 .IP \(bu 2
-\fB\fC\-K\fR \fIdistance\fP or \fB\fC\-\-cluster\-distance=\fR\fIdistance\fP: Cluster points (as with \fB\fC\-\-cluster\-densest\-as\-needed\fR, but without the experimental discovery process) that are approximately within \fIdistance\fP of each other. The units are tile coordinates within a nominally 256\-pixel tile, so the maximum value of 255 allows only one feature per tile. Values around 20 are probably appropriate for typical marker sizes.
+\fB\fC\-K\fR \fIdistance\fP or \fB\fC\-\-cluster\-distance=\fR\fIdistance\fP: Cluster points (as with \fB\fC\-\-cluster\-densest\-as\-needed\fR, but without the experimental discovery process) that are approximately within \fIdistance\fP of each other. The units are tile coordinates within a nominally 256\-pixel tile, so the maximum value of 255 allows only one feature per tile. Values around 10 are probably appropriate for typical marker sizes. See \fB\fC\-\-cluster\-densest\-as\-needed\fR below for behavior.
 .RE
 .SS Dropping a fraction of features to keep under tile size limits
 .RS
@@ -266,7 +269,7 @@ compensate for the larger marker, or \fB\fC\-Bf\fR\fInumber\fP to allow at most
 .IP \(bu 2
 \fB\fC\-pd\fR or \fB\fC\-\-force\-feature\-limit\fR: Dynamically drop some fraction of features from large tiles to keep them under the 500K size limit. It will probably look ugly at the tile boundaries. (This is like \fB\fC\-ad\fR but applies to each tile individually, not to the entire zoom level.) You probably don't want to use this.
 .IP \(bu 2
-\fB\fC\-aC\fR or \fB\fC\-\-cluster\-densest\-as\-needed\fR: If a tile is too large, try to reduce its size by increasing the minimum spacing between features, and leaving one placeholder feature from each group.  The remaining feature will be given a \fB\fC"cluster": true\fR attribute to indicate that it represents a cluster, a \fB\fC"point_count"\fR attribute to indicate the number of features that were clustered into it, and a \fB\fC"sqrt_point_count"\fR attribute to indicate the relative width of a feature to represent the cluster.
+\fB\fC\-aC\fR or \fB\fC\-\-cluster\-densest\-as\-needed\fR: If a tile is too large, try to reduce its size by increasing the minimum spacing between features, and leaving one placeholder feature from each group.  The remaining feature will be given a \fB\fC"cluster": true\fR attribute to indicate that it represents a cluster, a \fB\fC"point_count"\fR attribute to indicate the number of features that were clustered into it, and a \fB\fC"sqrt_point_count"\fR attribute to indicate the relative width of a feature to represent the cluster. If the features being clustered are points, the representative feature will be located at the average of the original points' locations; otherwise, one of the original features will be left as the representative.
 .RE
 .SS Dropping tightly overlapping features
 .RS
@@ -376,7 +379,7 @@ The postfilter receives the features at tile resolution, after simplification, c
 .PP
 The layer name is provided as part of the \fB\fCtippecanoe\fR element of the feature and must be passed through
 to keep the feature in its correct layer. In the case of the prefilter, the \fB\fCtippecanoe\fR element may also
-contain \fB\fCindex\fR, \fB\fCsequence\fR, and \fB\fCextent\fR elements, which must be passed through for internal operations like
+contain \fB\fCindex\fR, \fB\fCsequence\fR, \fB\fCextent\fR, and \fB\fCdropped\fR, elements, which must be passed through for internal operations like
 \fB\fC\-\-drop\-densest\-as\-needed\fR, \fB\fC\-\-drop\-smallest\-as\-needed\fR, and \fB\fC\-\-preserve\-input\-order\fR to work.
 .SS Examples:
 .RS