1 of 13

Review Session 2: Visualization

Yang LI

2021-09-24

2 of 13

Outline

  • What to convey through visualization
  • Pipeline to visualize X-space
  • Network style visualization of spaces
  • Extension materials

3 of 13

Why visualize the data?

  • You may never know some facts until you see the data

4 of 13

Why visualize the data?

  • You may never know some facts until you see the data
  • A good visualization enhance the understanding

White & Harary (2001)

Zachary (1977)

Perozzi, Bryan, et al (2014)

5 of 13

Focus on featured viz in Economic Complexity

  • Network visualization of X-spaces
  • UMAP visualization of X-spaces
  • Treemaps
  • Geographical Maps

​

General python visualization with matplotlib/seaborn/etc. could be learned at: https://colab.research.google.com/notebooks/charts.ipynb

6 of 13

Visualizing X-spaces as drawing a map

Use meaningful symbols to

  • Preserve the critical entities and relation/locations in space
  • Simplify the unnecessary part

​

7 of 13

Input data

  • Ecp/Rcp/Mcp: value/RCA/presence of region-products
  • Proximity matrix: relationship between entities
  • Feature of products (or industries/occupations/etc.):
    • Classification
    • Communities/clusters
    • PCI
    • Other statistics

8 of 13

Node elements for representation

Color of Nodes

Size of Nodes

A

B

Labels of Nodes

Location of Nodes

Shapes of Nodes

9 of 13

Edges are less used in representation

  • But it influence the aesthetics of network visualization greatly

Directed Edges

Labels on Edges

Thickness of Edges

Color of Edge

A

Shape of Edges

10 of 13

Pipeline to visualize X-space

  • Data collection and metric calculation (pre-sessions)
  • Extract the informative part of relation
  • Layout generation
  • (optional) cluster generation
  • Mapping desired property to visual elements
    • Whole space
    • Portfolio of regions

11 of 13

Network style visualization of spaces (colab)

  • Extract the informative part of relation: mst + high proximity
  • Layout generation: force-layout
  • (optional) cluster generation: community detection
  • Aesthetic mapping: desired property -> visual elements
    • Whole network
    • Portfolio

​

  • Interpretation & exploration of result

12 of 13

Extension materials

Alternative choices in the viz pipeline

  • Other transformations
    • Transform raw variable: log/pmi/binarization/etc.
    • (Dis)similarity metric: Euclidean/cosine/correlation/L*/divergence/…
  • Backbone extraction+layout algorithm:
    • KNN
    • UMAP/Tsne
    • Largevis
    • Other manifold learning algorithm

13 of 13

Extension materials

Alternative choices in the viz pipeline

  • Clustering tools:
    • other community detection algorithm such as infomap/walktrap/etc.
    • Dimension reduction + clustering algorithm
  • Plotting interface:
    • Gephi
    • Cytoscape
    • D3 and other javascript based libraries
      • And their python/jupyter interface

​