edxu96 / mgrid

multilayer network for power grid with multiple voltage levels

Edward has an MSc in engineering from Technical University of Denmark (DTU), and works as a mathematical modeller / software engineer at encoord in Essen, Germany. He is familiar with techniques from mathematical simulation, mathematical modelling and graph theory.

You can know more about him through https://edxu96.github.io/ and have a look at his resume.

mgrid's Issues

Convert planar graph to multilayer network

Test robustness of vertex split

To enhance #4.

One graph class for two operations

Only split method is tested.

Split one node in geometric graph

• some edge attribute should serve as cluster
• allow lambda function in test files

Init GitHub Actions

• https://github.com/snok/install-poetry

EU-LV test case

Ignore edge contraction for now

Attribute raw is not updated

Maintain an edgelist for now

• Not efficient.

No directions in components

The graph returned by Graph.with_cuts must preserve original directions.

Remove unnecessary methods and properties

• To reduce dependency.

Pass two vertex names directly

A dictionary for single to-be-split vertex. A data frame for multiple.

Utility to rename curve terminals

It is not necessary to assign every edge attribute in graph, because, for now, only one can be used in vertex splitting. Instead, a dictionary mapping new edges to the original after splitting can be used to update any dataframe associated with edges.

Method to compare with another set of vertices

• Not sure if it is necessary.
• Avoid storing all the vertices in a set.
• Compare number of vertices first.

Split node is not removed

Networkx graph as IO

No resulted new edge in vertex split

• Related to #4: split one node in geometric graph

Why operate on graph

• better validation
• resulted edges will divide graph into components, See #32

Subgraph with new edges being cut

Split multiple vertices

When to specify which vertices should be split, it is not recommended to
use some vertex attribute. Multiple sets of vertices can be passed.
Different arguments can be used each time accordingly.

Find cut(-set) between components

There can another graph never adding new edge from vertex splitting. Then we know how number of components grows. However, it cannot help find cut-sets.(not necessary, should have a property)

Components can be seen in different layers, and cut-sets connect them.

Steps

• Find all split vertex in first component.
• Find their pairs.
• If another component, the edge is in the cut-set.

Plot geometric graph in maps

Intro to multilayer network

To Do

• most general multilayer network
• supra-adjacency matrix
• nodes in different layers are usually distinguished

Inapplicable

• Tensorial formalism

Elaborate on why not node attribute

Init pre-commit

Merge columns in other dataframes

• Database-style DataFrame or named Series joining/merging

To Do

• for already-split vertices
• for original edges

Set the attribute of resulted edge None

Current, a new attribute split_ is set True. Instead, the attributed used in clustering associated edges can be set None, indicating that the attribute of the new edge is not meaningful.

Networkx function not work

Functions like edge_subgraph will not work if the class is unable to initiate empty networkx graph. The following TypeError will occur:

TypeError: __init__() missing 1 required positional argument: 'g'

A set for split vertex

• Union two sets: https://stackoverflow.com/a/49227886/10181743

What if graph is modified halfway

Use source-target instead of first-second

Split leaf vertex

• https://en.wikipedia.org/wiki/Vertex_(graph_theory)

Forget to remove original edge when renaming

Init class for planar graph

To Do

• find_inter_nodes
• planar_nodes
• from_edgelist
• handle isolated node

Gather split vertices directly

New naming function for two string with similar prefix

• Name the new node test when two labels are test_lv and test_hv.
• For edge contraction.

Collect resulted vertices in dataframe

Map vertex to its component in with_cuts

• Use networkx.node_connected_component(g, vertex)
• Components are found in with_cuts property. See #37: subgraph with new edges being cut

A case with multiple vertices to be split

Check if connected when calling property new

Redundant mapping in _renamed_dict

The final graph should be correct. But the correspondence between edges in the final graph and original edges has redundant values.

Why cut-related features

• Vertex-split-resulted edges are special.

Core Features

• with_cuts

New package name

Rename terminal of in-edges and out-edges

To enhance #4.

Numpy 1.19 not working

Use numpy 1.18.0 instead.

Subgraph of compoent

networkx.graph.subgraph(all_vertices_in_components)

Init geometric graph and contraction

• can be contracted
• new name for two terminals of contracted edge

The file conversion is not essential

Integrate contraction function in Graph.

Use directed graph

• Better correspondence
• If the graph is directed, g.edges(node) will not give the in-edges of node.
• Use pandas.DataFrame.update to update edgelist
• Graph.renamed_ is kept for now.

Import and specify logger in one file