Lab 04 - Entropy¶

Goal: calculate uncertainty and apply a routing policy.

Info: Entropy as safety valve

High entropy signals diffuse confidence. Red flag for auto-routing. Pair entropy thresholds with probability thresholds.

In [ ]:

import math

def entropy_bits(probabilities):
    return -sum(p * math.log2(p) for p in probabilities if p > 0)

peaked = [0.92, 0.05, 0.03]
flat = [0.45, 0.33, 0.22]
print("Entropy peaked:", round(entropy_bits(peaked), 4))
print("Entropy flat:", round(entropy_bits(flat), 4))

import math def entropy_bits(probabilities): return -sum(p * math.log2(p) for p in probabilities if p > 0) peaked = [0.92, 0.05, 0.03] flat = [0.45, 0.33, 0.22] print("Entropy peaked:", round(entropy_bits(peaked), 4)) print("Entropy flat:", round(entropy_bits(flat), 4))

In [ ]:

def normalized_entropy(probabilities):
    n = len(probabilities)
    return entropy_bits(probabilities) / math.log2(n)

for dist in [peaked, flat]:
    print([round(p, 2) for p in dist], "->", round(normalized_entropy(dist), 4))

def normalized_entropy(probabilities): n = len(probabilities) return entropy_bits(probabilities) / math.log2(n) for dist in [peaked, flat]: print([round(p, 2) for p in dist], "->", round(normalized_entropy(dist), 4))

In [ ]:

def route_ticket(probabilities, p_threshold=0.80, h_threshold=0.45):
    top_p = max(probabilities)
    h_norm = normalized_entropy(probabilities)
    return "auto-route" if (top_p >= p_threshold and h_norm <= h_threshold) else "human-triage"

print("Peaked policy:", route_ticket(peaked))
print("Flat policy:", route_ticket(flat))

def route_ticket(probabilities, p_threshold=0.80, h_threshold=0.45): top_p = max(probabilities) h_norm = normalized_entropy(probabilities) return "auto-route" if (top_p >= p_threshold and h_norm <= h_threshold) else "human-triage" print("Peaked policy:", route_ticket(peaked)) print("Flat policy:", route_ticket(flat))