Fit U-INVITE Network Demo

This demo illustrates how to estimate semantic networks using the U-INVITE and Hierarchical U-INVITE models — two of the most powerful and computationally intensive methods available in the SNAFU framework.

—

About U-INVITE:

  • U-INVITE uses a censored random walk model to infer latent semantic structure from fluency lists.

  • Estimation can be slow, especially with larger vocabularies or participant pools.

  • Hierarchical U-INVITE allows estimating both individual-level and group-level networks simultaneously.

  • You can also fit U-INVITE using a static prior (e.g., from USF norms) for improved speed and interpretability.

—

What This Demo Does:

  1. Loads animal fluency data (Experiment1) and prepares it for modeling.

  2. Sets up the DataModel (e.g., jump rates, censoring behavior) and Fitinfo (e.g., initialization strategy, priors).

  3. Runs three network estimation methods: - Example 1: Standard U-INVITE on a single participant’s fluency lists - Example 2: Hierarchical U-INVITE across multiple participants - Example 3: U-INVITE with a static prior (USF semantic network)

  4. Saves all estimated networks using pickle.

—

Key SNAFU Functions Used:

  • snafu.uinvite

  • snafu.hierarchicalUinvite

  • snafu.priorToGraph

  • snafu.genGraphPrior

  • snafu.load_network

Output Files:

  • uinvite_network1.pkl: Adjacency matrix for individual U-INVITE network

  • individual_graphs.pkl: List of participant-level networks (hierarchical)

  • group_network.pkl: Aggregated group network (hierarchical)

  • uinvite_network3.pkl: U-INVITE network using USF prior

Tips:

  • Consider reducing prune_limit, triangle_limit, and other_limit in fitinfo to reduce runtime.

  • You can use a pre-defined network (like USF) as a prior for faster, guided fitting.