mathematica

Computing $p$-values using Mathematica

Preliminary

As described in ../../README.md, you must first run ../Table1to3_make_ultraviolet_and_light_embeddings_csv.py.

This script outputs the component values (in CSV format) of the normalized embeddings of "ultraviolet", "light", and "ultraviolet $\odot$ light" for GloVe, GPT-2, and Pythia-160m. As an example, the output used in the paper is placed under data.

Run

product_of_two_normal.nb is a Mathematica notebook for computing $p$-values for the component values of these vectors. In particular, it shows the computation results for GloVe. To do the same for GPT-2 or Pythia-160m, simply change the path to the input CSV file. A PDF rendering of the notebook is provided in product_of_two_normal.pdf.

The experimental results for GloVe, GPT-2, and Pythia-160m are available in ../../data/ultraviolet_and_light_pvalues.