INFORMATION
Shannon entropy
Quantifies uncertainty in a probability distribution.
Example of usage
Three customer service topics have 1.5 bits of uncertainty.
Usage / limitations
+Use when: Measure uncertainty, information content and diversity.
−Limitations: Requires a valid probability distribution; zero probabilities contribute zero.
Mathematical formula
H(X) = -Σ p(x) log₂ p(x)Sample dataShow
{
"variable": "support_tickets",
"domain": "Customer support triage",
"feature_names": [
"issue_type",
"probability"
],
"records": [
{
"issue_type": "billing",
"probability": 0.5
},
{
"issue_type": "delivery",
"probability": 0.25
},
{
"issue_type": "returns",
"probability": 0.25
}
]
}Code example (Python)
# DataLens.ai sample data — customer support ticket categories
support_tickets = [{'issue_type': 'billing', 'probability': 0.5}, {'issue_type': 'delivery', 'probability': 0.25}, {'issue_type': 'returns', 'probability': 0.25}]
# DataLens.ai example
# Compute the uncertainty of the support issue distribution in bits.
from math import log2
probabilities = [row["probability"] for row in support_tickets]
assert abs(sum(probabilities) - 1) < 1e-12
entropy_bits = -sum(p * log2(p) for p in probabilities if p > 0)
assert abs(entropy_bits - 1.5) < 1e-12
print("Expected output verified")
# DataLens.ai example
# Compute the uncertainty of the support issue distribution in bits.
from math import log2
probabilities = [row["probability"] for row in support_tickets]
assert abs(sum(probabilities) - 1) < 1e-12
entropy_bits = -sum(p * log2(p) for p in probabilities if p > 0)
assert abs(entropy_bits - 1.5) < 1e-12
print("Expected output verified")