Lab 06 - Deterministic vs Stochastic¶

Goal: compare argmax routing with probabilistic sampling.

Info: Seeding for reproducibility

In compliance workflows, determinism is mandatory. Always seed random behavior for audit trails and incident investigation.

Info: Argmax vs sampling

Argmax is deterministic and fast; sampling is stochastic and diverse. Choose per decision: argmax for routing, sampling for drafts.

In [ ]:

import random

intents = ["account_unlock", "vpn_issue", "security_incident"]
probs = [0.58, 0.31, 0.11]

def argmax_choice(labels, p):
    idx = max(range(len(p)), key=lambda i: p[i])
    return labels[idx]

print("Deterministic choice:", argmax_choice(intents, probs))

import random intents = ["account_unlock", "vpn_issue", "security_incident"] probs = [0.58, 0.31, 0.11] def argmax_choice(labels, p): idx = max(range(len(p)), key=lambda i: p[i]) return labels[idx] print("Deterministic choice:", argmax_choice(intents, probs))

In [ ]:

def sample_choice(labels, p, seed=None):
    rng = random.Random(seed)
    r = rng.random()
    total = 0.0
    for label, pr in zip(labels, p):
        total += pr
        if r <= total:
            return label
    return labels[-1]

print("Seeded run 1:", sample_choice(intents, probs, seed=7))
print("Seeded run 2:", sample_choice(intents, probs, seed=7))
print("Unseeded run:", sample_choice(intents, probs))

def sample_choice(labels, p, seed=None): rng = random.Random(seed) r = rng.random() total = 0.0 for label, pr in zip(labels, p): total += pr if r <= total: return label return labels[-1] print("Seeded run 1:", sample_choice(intents, probs, seed=7)) print("Seeded run 2:", sample_choice(intents, probs, seed=7)) print("Unseeded run:", sample_choice(intents, probs))