MediumNLPnlp-120

PCFG Rule Probability Normalization

Problem

Normalize rule counts into PCFG probabilities per left-hand side. Keys are rules formatted as 'LHS -> RHS' and values are their counts.

For each rule return count divided by the total count of all rules sharing the same LHS, so every LHS distribution sums to 1.

Hint

The left-hand side is the first space-separated token of the rule string.

case 1

in: [{"S -> A B":3,"S -> C":1,"A -> a":2}]

expected: {"S -> A B":0.75,"S -> C":0.25,"A -> a":1}

case 2

in: [{}]

expected: {}

case 3

in: [{"X -> y":5}]

expected: {"X -> y":1}

case 4

in: [{"S -> A":2,"S -> B":2,"T -> c":1}]

expected: {"S -> A":0.5,"S -> B":0.5,"T -> c":1}

Starter codePython

def pcfg_normalize(counts):
    # Your code here
    pass

4 test cases · run your code against them in the editor