Compare Networks Demo

This demo illustrates how to simulate and compare semantic fluency data using two different networks — the original USF network and a perturbed version of it. This is useful for evaluating how well a model-generated or alternate network captures actual fluency behavior.

Overview:

  1. Load a known semantic network (USF animal subset).

  2. Create a perturbed version of the network by randomly flipping ~10% of edges.

  3. Generate fluency lists from both networks using censored random walks.

  4. Compute log-likelihoods of those lists under both networks.

  5. Save likelihoods and data for further inspection.

Key Concepts:

  • Perturbation: Randomly flip edges to simulate variation or error in network structure.

  • Fluency Simulation: Generate fluency lists using random walks with optional jump probabilities.

  • Likelihood Comparison: Assess how well each network explains data generated from itself and the other network.

Functions Used:

  • snafu.read_graph

  • snafu.DataModel

  • snafu.gen_lists

  • snafu.probX

Output:

  • Prints log-likelihoods to console

  • Saves rounded log-likelihoods in demos_data/expected_likelihoods.pkl