Naive Bayes: Subtleties

2.

what if none of the training instances with target value 112#112 have attribute value 115#115? Then

127#127

128#128

Typical solution is Bayesian estimate for 129#129

130#130

where

• 131#131 is number of training examples for which 132#132,
• 133#133 number of examples for which 132#132 and 134#134
• 76#76 is prior estimate for 129#129
• 135#135 is weight given to prior (i.e. number of ``virtual'' examples)