4 ms·
That N/p rule of thumb is ultimately subjective, except for the fact that when N/p < 1, you can't use OLS anymore (because there is an infinitude of zero-traini
by michaelochurch 12y ago
That N/p rule of thumb is ultimately subjective, except for the fact that when N/p < 1, you can't use OLS anymore (because there is an infinitude of zero-training-error solutions). It depends on correlations of predictors and signal/noise ratio as well. Those are just approximate guidelines.