himnabil/dojo-naive-bayes-redis

dojo-naive-bayes-redis

domain

naive bayes

p(category given observation) = 
    p(category) * p(observation given category) / p(observation)  

p(observation given category) = 
    let p = 1 for feature in observation : p *= p(feature given category)

p(observation) = 
    let p = 1 for feature in observation: p *= p(feature)


p(feature) = 
    nb(observation with feature) / nb (all observation)

p(category) = 
    nb(observation labeled category) / nb (all observation)

p(featue given category) = 
    nb(observation with feature, labeled category) / nb (observation labeled category)

a negation of feature (!feature)

p(!feature) = 1 - p(feature)
if (feature in not observation) => feature in not observation   

backlog

1. reporting a labeled observation to a naive bayes model instance
2. invoking a model instance to get the probabilities of each category given an input unlabeled observation
3. capability to scale (supporting load)
4. getting a metadata for model instance
5. reporting a unlabeled observation to a naive bayes model instance then set label later

Constraint

• TDD ( unit test ) no integration test needed
• Pear programing or mob programing
• using REST api
• storing date in memory for US 1 and 2, and a key value db (redis like) for US 3,4,and 5