We noticed floating point rounding errors were masking distinction between very high p-values that were in fact different. The method now calculates and works with the log p-values internally, and makes them available via the `all_log_p_values_` attribute.
Due to the problems with the un-logged p-values, `all_p_values_` is deprecated, and will be removed in v3.0. We encourage all users to avoid using it, and instead exponentiate `all_log_p_values_` (with care!) as needed.
To help understand the significance of different p and voter thresholds, we have added `plot.threshold` to provide a visualization of their interaction.