Adversarial Search
  - brancing factor
  - utility
  - state of the art, chess, machine learning, pruning
  - inefficiencies
  - optimal moves
  - minimax, initial state, successor, terminal test/states, utility function
  - alpha-beta pruning
  - cut-off test (depth limit quiescence search), terminal test, eval vs utility
  - nondeterminism - expectiminimax


Mancala
 --results/implementation


Probability
  - probability 
     - # appeared / # trials
  - Total Probability = sum P(A|xi) * P (xi)
  - Bayes Rule
     - P(A|B) = P(B|A) * P(A) / P(B)
  - smoothing
     - ( # appeared + lambda ) / (#totals+#classes*lambda)
     - laplacian smoothing - lambda = 1
     - laplacian smoothing with two classes -> (#appeared+1)/(#totals+2)
  - probability of spam
     - product ( P(words) ) / ( product( P(words) ) + product(1 - P(words)) )