Calculating metrics using filtered data

Nov 8, 2010 at 5:51 PM

Hello,

I am finding that I can do some pretty amazing visual analysis by using the dynamic and normal filters.  However, I believe I am observing that the metrics are always calculated on the complete data set.  Is this correct?   The group I am analyzing is broken up into teams, and I would like to determine the centrality measures within each team individually.  Since I can display the individual teams (as well as differentiate them by shape or color), it seems to me that this should be possible.  Is there a way?

Thank you.

Peter

 

Coordinator
Nov 8, 2010 at 8:13 PM

Thank you for the interest in NodeXL.  

NodeXL does allow you to calculate metrics within groups of vertices.  Use the group features to collect sets of vertices into a group.  Then use the feature in Graph Metrics to calculate metrics within groups.  Metrics for each set of vertices will appear on the Groups worksheet in the NodeXL workbook.

Regards,

Marc

Nov 10, 2010 at 4:50 PM
Hi Marc,

Thank you for your reply. I don't know what you mean by group features.
I know you can ask NodeXL to find groups, but my teams are highly
interconnected, and cannot be identified by NodeXL as distinct entities.
Is there another way to identify subgroups, manually perhaps? I know who
is on what team and can display them seperately, so I am asking if there
is some way that I can get the metrics based on just what is being
displayed? It would be a very useful feature.

Thanks.

Peter

On Mon, November 8, 2010 4:13 pm, marcsmith wrote:
> From: marcsmith
>
>
> Thank you for the interest in NodeXL. NodeXL does allow you to calculate
> metrics within groups of vertices. Use the group features to collect
> sets of vertices into a group. Then use the feature in Graph Metrics to
> calculate metrics within groups. Metrics for each set of vertices will
> appear on the Groups worksheet in the NodeXL workbook.Regards,Marc
>
>
Coordinator
Nov 11, 2010 at 6:31 AM

Hello!

NodeXL has features to support the collection of vertices into "Groups".  You may want to consider upgrading your copy of NodeXL to the latest to ensure that you have all of these features.

Groups can be constructed automatically using the "Group" feature in the NodeXL menu.  You may select one of three different methods of creating a cluster of vertices from the "Cluster Options" menu.  Your choices are Clauset-Newman-Moore, Wakita and Tsurumi, and Girvan-Newman.  When you select the algorithm that suits your needs, use the "Find Clusters" command to create the collections of vertices, which then appear in the "Groups" worksheet.

Groups can be created manually using the "Group by Vertex Attribute" feature.  This allows you to identify each vertex that should be in a particular group based on an existing column of data in the Vertex worksheet.  You can add a new column to this worksheet and manually assign a value to each Vertex to place it in the group you define.

When groups are used, the "Graph Metrics" feature can be used to create network metrics for each cluster of vertices.  These metrics are calculated the same way that overall graph metrics are calculated, except all vertices not in the group are ignored.  The same procedure is followed for every group.

Regards,

Marc

Nov 24, 2010 at 8:49 PM
Hi Marc,

Thank you for the suggestions. I now know how to make my own groups, a
feature which had escaped me before, and I can display them individually.
This proves to be very useful.

However, I am still unable to get the metrics of the individual groups as
entities unto themselves The metrics that are calculated are still the
same as those for the entire data set. For instance, if a node x has a
degree of N within the whole data set (of a connected graph), NodeXL still
calculates that degree as N, even if it has only a degree of M taking into
account only the nodes in the group, even if I am only displaying that
group, with all other groups being filtered out.

Is there some way to get NodeXL to calculate the metrics of individual
groups as I define them and as they are displayed?, ignoring their ties to
the other nodes in the data set.

Thank you .

Peter


On Thu, November 11, 2010 2:31 am, marcsmith wrote:
> From: marcsmith
>
>
> Hello!NodeXL has features to support the collection of vertices into
> "Groups". You may want to consider upgrading your copy of NodeXL to the
> latest to ensure that you have all of these features.Groups can be
> constructed automatically using the "Group" feature in the NodeXL menu.
> You may select one of three different methods of creating a cluster of
> vertices from the "Cluster Options" menu. Your choices are
> Clauset-Newman-Moore, Wakita and Tsurumi, and Girvan-Newman. When you
> select the algorithm that suits your needs, use the "Find Clusters"
> command to create the collections of vertices, which then appear in the
> "Groups" worksheet.Groups can be created manually using the "Group by
> Vertex Attribute" feature. This allows you to identify each vertex that
> should be in a particular group based on an existing column of data in
> the Vertex worksheet. You can add a new column to this worksheet and
> manually assign a value to each Vertex to place it in the group you
> define.When groups are used, the "Graph Metrics" feature can be used to
> create network metrics for each cluster of vertices. These metrics are
> calculated the same way that overall graph metrics are calculated, except
> all vertices not in the group are ignored. The same procedure is
> followed for every group.Regards,Marc
>
>
Nov 24, 2010 at 8:58 PM

Peter:

The metrics that you are looking at on the Vertices worksheet, including Degree, are computed for the graph as a whole, regardless of how you have grouped your vertices.

Marc was talking about what we call "group metrics."  If you go to NodeXL, Analysis, Graph Metrics and check "Group metrics," then NodeXL will look at each of your groups as if they were disconnected from the rest of the graph, compute group metrics for those groups, and store the results in the Groups worksheet.  It's that last part that I think you're missing.

The group metrics are the same metrics we calculate when you check "Overall metrics": vertex counts, edge counts, and so on.  We do not computer per-vertex metrics, such as Degree, for the groups.

-- Tony

Nov 28, 2010 at 7:50 PM
Hi Tony,

Thank you very much. That's the information I needed. :-) What are the
chances of sometime in the future the centrality measures being calculated
for just the vertices that are visible on the graph? I imagine that
others would find that useful, as well.

For now, I am filtering my data in another spreadsheet and importing each
group as a separate entity.

Peter

On Wed, November 24, 2010 4:59 pm, tcap479 wrote:
> From: tcap479
>
>
> Peter:The metrics that you are looking at on the Vertices worksheet,
> including Degree, are computed for the graph as a whole, regardless of
> how you have grouped your vertices.Marc was talking about what we call
> "group metrics." If you go to NodeXL, Analysis, Graph Metrics and check
> "Group metrics," then NodeXL will look at each of your groups as if they
> were disconnected from the rest of the graph, compute group metrics for
> those groups, and store the results in the Groups worksheet. It's that
> last part that I think you're missing.The group metrics are the same
> metrics we calculate when you check "Overall metrics": vertex counts,
> edge counts, and so on. We do not computer per-vertex metrics, such as
> Degree, for the groups.-- Tony
>
>
>
Nov 29, 2010 at 3:33 PM

Peter:

I've added "per-vertex group metrics" to our feature request list.  Thanks for letting us know that it would be useful; we often have to guess about such things when deciding which features to implement.

-- Tony