I have the definition of an object provided as features probability. Each object has it's own feature importance and probabilities. For example for object "X", I have "color" feature (with the weight of 0.8) - the object can be blue in 80% of cases and black in 20% of cases. And "shape" feature (with the weight of 20%) - square in 30% and round in 70%.
I'm trying to create a "predictor", so if I'm observing something blue and round - (0.8 x 0.8) x (0.2 x 0.7) - probability for object X.
- Does it make any sense mathematically?
- If this method sounds reasonable enough, how should I handle really small numbers (I can have a really long vector of features, the final number will be really small)?