Zero-shot Visual Reasoning through Probabilistic Analogical Mapping

Analysis code and behavioral data for the paper Zero-shot Visual Reasoning through Probabilistic Analogical Mapping.

The following command will reproduce Figure 5 (comparing human 3D object mapping with visiPAM), and perform repeated-measures ANOVA:

python3 ./analysis.py

The corresponding data can be found in human_visiPAM_data.csv.

To perform the version of the analysis in which human behavioral responses are clustered (to account for bimodal response distributions), run:

python3 ./analysis.py --diptest

To perform correlation analysis (item-level correlation between human and visiPAM responses), run:

python3 ./correlation_analyses.py

To perform correlation analysis with the node-only visiPAM ablation model, run:

python3 ./correlation_analyses.py --model node_only

To perform correlation analysis with the edge-only visiPAM ablation model, run:

python3 ./correlation_analyses.py --model edge_only

Raw human behavioral data (including marker placement for each trial) can be accessed in the file raw_data.human.xlsx.

Behavioral stimuli can be found in the directory ./stimuli.

Pointcloud data can be found in the directory ./pointclouds.

Prerequisites

• Python 3
• NumPy
• SciPy
• Matplotlib
• pandas
• pingouin

Authorship

All code was written by Taylor Webb and Shuhao Fu.