An introduction to graph analytics using Gruff: a Visual Discovery Tool by Jans Aasman

1. GRUFF
Becoming a power user
San Diego Version
Jans Aasman, Ken Cheetham
Franz Inc

2. Highlights
• Navigate large graphs in a user friendly way
• Automatically derive SPARQL and Prolog queries from
interesting patterns on the screen.
• Discover links between nodes automatically
• Use a real AI language to explore the graph (Prolog)
• And some other goodies
– Programmatically control Gruff over http
– Show pictures for nodes.

3. 11 short lessons
1. Download the right Gruff (standalone vs server version)
2. Manage triple stores
– Import triples, display triples, freetext index triples.
3. Graph view: see triples as nodes and links
4. Table view: see triples as objects in tables
5. Outline view: see triples as hierarchies
6. Query view: writing SPARQL and Prolog for real men
7. Visual Query builder: build queries using point and click
8. Visual Discovery: build queries using patterns on the screen
9. Control Gruff through REST
10. Use pictures on nodes
11. Server edition of AllegroGraph and Gruff.

4. [1] Download the right Gruff
Franz provides two version of Gruff
• Version 5 for AG 33
– Standalone, AllegroGraph built in.
– Works on all platforms: Windows, Linux, Mac, 32 bit and 64 bit.
– Perfect for exploring small and medium sized datasets
• Version 5 for AG 4.10
– For very large datasets and production systems
– If you need an enterprise database with transactions and ACID
behaviour, backup, restore, point in time recovery, warm failover,
triple level security, server side javascript etc etc.
– Windows/Mac/Linux client for AllegroGraph server edition
– And if you need SPARQL 1.1 and SPIN

5. My demo today

6. [2] Manage your triple store
• Create a new triple store (file menu)
– Make sure you fill in your machine name for local triple
stores
• Load triples from a file or the web (file menu)
• Display all triples up to a limit (display -> display all triples up to a limit)
• Remove all triples from screen (‘z’ till blank, or ctr-alt-x)
• Create free text index (display -> edit freetext predicates)
• Search for something using freetext (;)

7. Oct 2007

10. Clinical
Trials
Brand
Medicin
Drugs
Side
Effects
Diseases
SNOMED MESH
Patient
Records

11. Clinical
Trials
Brand
Medicin
Drugs
Side
Effects
Diseases
SNOMED MESH
Patient
Records
Systematic Nomenclature of Medicine –
Clinical Terms
National Library of Medicine's controlled
vocabulary thesaurus
Records of Patients
tripping and falling

12. [3] Graph View ‘G’
• The view you’ll spend 80 % of your time in (well, I do)
• Some easy starters to get something on the screen
– Using freetext indexing (;)
– Display all triples up to a limit (display menu)
– Do simple SPARQL query in query screen (w) .. Select * where { ?x a ?y . } limit 10 ..
Click on ‘create-visual-graph’
– Display triples by type (display menu)
• Select the predicates to show on the screen (p)
• Explode nodes (f)
• Going back to previous state (z)
• Reorder graph ( r )
• Nudge the graph (d)
• Zoom in and out (middle wheel mouse, or use , and .)

13. [3] Graph View ‘G’ continued
• Explode node by right-clicking and choosing from a menu or a tree
• Delete nodes from the screen (x)
• Delete all nodes from the screen (ctr-alt-x)
• Eraser mode (ctr-shift-x)
• Remove orphans (shift-o)
• Show full namespaces (shift-8 .. Not F8)
• Show predicate names on screen (n)
• Find shortest path (shift-f)
– Pin nodes, or graph might change too much (i)
– What if you can’t find a path? Choose more predicates
and sometimes less 
– Create larger time window to search
– Buy more memory and solid state drives for your laptop
– Check in table view if one of the nodes has the right predicates

14. [3] Graph View ‘G’ continued
• Sometimes you don’t want to see the node labels (ctr-8 or
– Visual-graph-options -> node labels -> use label preds for nodes)
• You can add your own node labels.. Very useful
• You can always see the full URL of a node left under the screen
• Double click on node to go to table view (or hit ‘t’)
• Go directly to outline view from selected node (o)
• Save a graph layout to disk (ctr-s)
• Load a graph layout (ctr-l)
• Graph layouts are simple text files and can be shared

15. [4] Table View ‘T’
IF YOU WANT TO SEE YOUR TRIPLES IN A TRADITIONAL TEXTUAL TABLE
• Navigate the data by clicking on ‘values’
• Above gray line: s p o (where s is the main node we selected)
• Under gray line: o p s (note the ‘is’ predicate ‘of) to denote
the other direction
• Note how ‘8’ shows you full URLs of nodes
• Note how ‘Alt-8’ shows you how we use node-labels
• You can create triples in the table view, see manual…

16. [5] Outline View ‘O’
IF YOU HAVE DEEP HIERARCHICAL DATA AND YOU NEED TO RETAIN THE CONTEXT
WHILE NAVIGATING.
• Navigate the data by clicking on little arrows
– Blue: outgoing triples
– Black: triples pointing to previous level (not the ‘is’ … ‘of’
around the predicates
• Display:
– Note how ‘8’ shows you full URLs of nodes
– Note how ‘Alt-8’ shows you how we use node-labels
• You can create triples in the table view, see manual…

17. [6] Query View – ‘W’
• Write queries in SPARQL and Prolog – by writing code
• Use namespace abbreviations (b) to avoid typing prefixes
• Turn output into Graph or CSV
• Managing queries
– Save queries (ctr-s)
– Load queries (ctr-l)
• Investigate menu … query options
– One can change the time limit..

18. [7] Visual Query Editor ‘E’
• Build queries visually – no code!
• Most things work like the graphical query view
– Search for freetext (;)
• Right click on screen to create new variables
• Right click on variables to create new predicates with drag
and drop
– Multiple ways to choose names of predicates.. See menus

19. [8] Discovery: using patterns on
the screen to build queries
• Create interesting patterns on the screen using everything
you learned so far
• Find a trail that you think is interesting
• Highlight each node in trail (ctr-left-click or ‘h’)
• Move highlighted nodes to query editor (ctr-alt-h)
• Change some variables

20. [9 ]Controlling Gruff
programmatically through HTTP
• Start Gruff as a server:
– Global options -> communications -> start http:server
• Documentation:
– F1 in Gruff, search for http
• Two very simple examples:
– http://localhost:8000/update-layout?redo=yes
– http://localhost:8000/store-info?attribute=store-name
• And a slightly more complicated one: finding links between two nodes
• http://localhost:8000/find-
paths?node1=http%3A%2F%2Fwww.cyc.com%2F2002%2F04%2F08%2Fcy
c%23economicStability&node2=http%3A%2F%2Fwww.cyc.com%2F2002%
2F04%2F08%2Fcyc%23CoalitionPartnerFieldOPSSecretLAN-
DASSA&predicates=current&keep-old-nodes=yes

21. [10] Using Pictures for Nodes

22. [10] Using Pictures for Nodes
• Tell Gruff what predicate to use for pixmaps
– <jans> <pixmap> “pictureOfJans.ico”
– Global-options -> node label predicates -> custom predicates for node label pixmaps.
• And tell Gruff the base location of your pixmaps
– You can use relative filenames
– You can use jpg png bmp ico
– Global-options->miscellaneous->document base folder

23. [11] Using Gruff with the Server
• First: install Server, or VM or use Hosted version…

24. [11] Using Gruff with the Server
• Second: download Gruff client for Server

25. In this example I start the VM
• Click on agstart.sh

26. • Find the ip address of the server (if you don’t have fixed ip)

27. Road Map
• Connecting with other triple stores
• Circle Views, Hierarchical Views, etc
• Statistical Analysis and Business Intelligence functionality
• Bayesian belief networks
• Extensive support for GeoSpatial, Temporal and Social Network Analysis in
SPARQL
• Tell us What You Want.