NTDS'19 projects grading

November 6, 2019 ยท View on GitHub

Below is a detailed description of the five criteria on which projects will be evaluated. Bonus points can be attributed in each category for going the extra mile.

1. Story (20 points)

  • motivation: why study this question or develop that product?
  • relevance of chosen data and tools
    • Can the data answer the question or support the product?
    • Are the chosen tools relevant?

2. Acquisition (10 points)

  • proper data: it has a graph structure, and nodes have attributes (e.g., features, labels, time series)
  • example bonus: collection, processing, and cleaning of primary or complementary data (e.g., using a web API, combining datasets, designing a scheme to weight the edges)

3. Exploration (20 points)

  • some properties of the graph (e.g., connected components, sparsity, diameter, clusters, degree distribution, spectrum)
  • identify the type of graph (e.g., power law, small world, regular, sampled manifold)
  • some properties of the nodes (e.g., clustering coefficient, modularity, centrality)
  • some analysis of the attributes (e.g., their distribution, smoothness, graph Fourier transform)
  • a vizualization of the network
  • a reflection on the insights

4. Exploitation (30 points)

  • at least one of the following tools, seen during the lectures, is used:
    • clustering (spectral clustering, k-means)
    • graph Fourier transform
    • regularization (graph Tikhonov, graph total variation)
    • dimensionality reduction (PCA, MDS, LLE, ISOMAP, Laplacian eigenmaps, t-SNE)
    • graph filters (Chebyshev, ARMA)
    • graph neural networks
  • critical evaluation of the results
    • subjective or objective (baseline, existing work)
    • state the limitations: to what extent did you answer the question or provide a good product
  • example bonus: use multiple relevant tools, or tools beyond what was seen in class

5. Communication (20 points)

  • report:
    • structure
    • content: be explicit about what you did in acquisition, exploration, exploitation. It should be precise enough for another team to replicate your work. Results should be supported by figures, tables, etc.
    • well written: clarity, conciseness
  • oral: presentation skills (organization, clarity)
  • github repository: README, LICENSE, documented, organization (notebooks and python modules), reproducible, good coding practice (comments, docstrings)
  • example bonus: published interactive visualization