3 ms·
"Naïve Bayes can only represent non-negative frequency counts of features; therefore it was not a candidate as accelerometer data has negative values. However,
by phezzan 10y ago
"Naïve Bayes can only represent non-negative frequency counts of features; therefore it was not a candidate as accelerometer data has negative values. However, this could be mediated by simply scaling all the data to ensure positive values (i.e. multiplying all values times 100)."
I think there are two serious flaws here.
- Bayesian frequency counts aren't measured values - they're counts of measured values...
- Multiplying a dataset of positive and negative numbers isn't going to make it strictly positive (unless you multiply by 0). You'd have to add the minimum value to all values.
This lead me to look for the author, but sadly the author is some anonymous 'GuestBlogger' ...
- garborg 10y agoSecond flaw is even better than that -- author mistakenly asserts 'multiplying all values times 100' is the _only_ way to 'scale all data to ensure positive values.'