Skip to main content

Accuracy vs GM prevalence

Created on June 30|Last edited on June 30
This report shows the tradeoff between accuracy and GM prevalence for randomly generated datasets. Each scenario is a different way of generating random datasets.
The scatter plots are interactive, meaning that you can click on a point and you will be directed to the associated run. Each run corresponds to a specific dataset (that was randomly initialised) and trained from 100 different random weight initialisations.
Below each chart, there are two options:
  • "all" displays all runs
  • "with GM" displays only those runs where the GM was reached at least once during the 100 training trials

Scenario #1: 7 red, 5 blue, 3 hyperplanes


0.00.10.20.30.40.50.60.70.80.91.0gm_prevalence0.840.860.880.90.920.940.960.981acc_mean
all
264
with GM
264


Scenario #2: 7 red, 5 blue, 2 hyperplanes


all
90
with GM
90