SciGraphs · Import an abstract graph
Turn a plain edge-list CSV into the Blender graph object every other SciGraphs panel operates on. An abstract graph brings no coordinates, so this walkthrough also decides where its nodes go.
Mirrors the Data panel.
Before you begin
Open the 3D Viewport sidebar with N, pick the SciGraphs tab and expand Data. Have an edge list ready as a CSV; anything with two identifier columns works. This page uses a six-line friends.csv:
source,target,strength
ana,bruno,3
ana,carla,1
bruno,carla,2
carla,diego,5
diego,elena,1
elena,ana,41. Load the file
Leave Step 1: Select Data Source on File; the alternatives are Database (a live SELECT), SuiteSparse (a matrix fetched by identifier) and Repro (replaying a saved pipeline).
Set the Filepath, leave Delimiter at ,, and press Load File. The panel reads the header, tags each column as numeric, datetime or text, and unlocks Step 2. A GEXF file needs none of this: the panel reports Native graph file detected and maps the structure itself.
Load File guesses whether your data is geographic from column names, by substring. The latitude patterns include a bare y and the longitude patterns a bare x, so ordinary numeric columns like density, entropy, delay, max_load, index or flux match. When one of each hits, the Geospatial Options checkbox switches itself on and stays on.
That is not cosmetic. Create Graph then takes the geospatial branch: it builds an Earth globe, and because Auto-Geocode Countries is on by default it sends your node identifiers to Nominatim as if they were place names. After every Load File, glance at the Geospatial Options header and untick it unless you actually want a globe. Notebook 01 reproduces the trap with two innocuous column names.
2. Define the graph structure
In Step 2: Define Graph Structure, set Source Nodes to source and Target Nodes to target. Two switches here change what later panels will let you do:
- Directed Graph unlocks the directed analyses and the flow panels, and changes what the hierarchical layouts can find. It is fixed at import time.
- Remove Self-Loops should stay on unless a node pointing at itself is meaningful in your data.
SciGraphs stores the node and edge lists as one comma-joined string on the object, and splits them back on commas. A name like Smith, J. becomes two identifiers and shifts every identifier after it. With the automatic layout on, the import dies with an IndexError raised from inside the layout, several steps away from the cause. Rename or quote such identifiers out of the CSV first; integers and bare words are always safe.
3. Pick the attributes to import
Set Weight Column to strength if you want it treated as an edge weight, then use the Import as Attributes checklist for everything else. Only ticked columns reach the mesh, so tick what you intend to color or size by later.
4. Create the graph object
Step 3: Create Graph Object carries one decision, Auto Layout on Import, and it also decides the order of the rest of your session. Leave it on for a first look and the graph is created and immediately placed with the Default Layout (ForceAtlas2) at the given Scale. Turn it off if you are importing edge-valued attributes you intend to keep, for the reason below.
Press Create Graph. The object appears in the viewport and the Layout panel’s status card starts reporting its node and edge counts.
Applying a layout, including the automatic one on import, calls rebuild_edges, which deletes and recreates every mesh edge. Attributes on the POINT domain survive; attributes on the EDGE domain come back allocated, correctly named, the right length, and filled with zeros. Nothing raises, and the giveaway is a figure colored by an edge attribute that renders as one flat color.
So the safe order is the opposite of the natural one: import with Auto Layout on Import off, lay the graph out in the Layout panel, then compute or import anything edge-valued. Notebook 01 measures this on a real column, including the surviving POINT aggregates of the same data.
5. Attach a node attribute table
If your measurements live in a second file, keep the graph active and use Import Node Attributes. Rows are matched to nodes by name in the form node_name <sep> value(s); any node missing from the file gets NaN rather than a silent zero. This route is on the node domain, so the layout warning above does not apply to it.
6. Make it renderable
Step 4: Apply Visualization runs Setup Visual, which adds the Geometry Nodes setup that draws node spheres and edge cylinders. Until you press it the object is topology only and renders as bare mesh.
Switch Step 1 to SuiteSparse, type a Matrix Identifier such as Grund/bayer09 and press Download & Create Graph. Graph Representation chooses between a symmetric adjacency and a bipartite reading, and Giant Component Only discards stray fragments.
An edge list cannot express an isolated node. A node with no edges has no row to appear in, so it is absent after import, and the File source has no matrix mode: a square matrix handed to it is read as an edge list between its first two columns. Use SuiteSparse for matrix data, or add a placeholder edge and delete it afterwards.
There is no seed field anywhere in this panel. With Auto Layout on Import on and a Spring algorithm selected, two identical imports of the same file produce two different drawings. The Layout tutorial explains which algorithms are affected, and Reproducible pipelines covers the route that does give a fixed answer.
Next steps
Lay out a graph decides where the nodes go, which this panel only picked a default for.