Audit trails for model decisions in MLOps - Mini Project: Build & Apply

Audit Trails for Model Decisions

📖 Scenario: You work in a team that builds machine learning models. Your team wants to keep a clear record of every decision the model makes. This helps to check and understand the model's behavior later, like keeping a diary of its choices.

🎯 Goal: Build a simple program that logs each model decision with details like input data, prediction, and timestamp. This log acts as an audit trail to track model decisions over time.

📋 What You'll Learn

Create a list to store audit trail entries

Add a configuration variable for the model name

Write a function to log decisions with input, prediction, and timestamp

Print the audit trail entries

💡 Why This Matters

🌍 Real World

Audit trails help teams track and review machine learning model decisions. This is important for debugging, compliance, and improving trust in AI systems.

💼 Career

Knowing how to implement audit trails is useful for roles in MLOps, data science, and software engineering where model transparency and accountability are required.

1

Create the audit trail list

Create a list called audit_trail that will hold all the model decision records.

MLOps

# Create a list called audit_trail
# Your code here

Need a hint?

Use square brackets [] to create an empty list named audit_trail.

2

Add model name configuration

Create a variable called model_name and set it to the string "SimpleClassifierV1".

MLOps

audit_trail = []
# Create a variable model_name with value "SimpleClassifierV1"
# Your code here

Need a hint?

Use an equals sign = to assign the string "SimpleClassifierV1" to model_name.

3

Write the logging function

Write a function called log_decision that takes input_data and prediction as parameters. Inside the function, import datetime, get the current time with datetime.datetime.now(), and append a dictionary with keys model, input, prediction, and timestamp to the audit_trail list.

MLOps

audit_trail = []
model_name = "SimpleClassifierV1"
# Define function log_decision(input_data, prediction)
# Your code here

Need a hint?

Remember to import datetime inside the function and use audit_trail.append() to add the record.

4

Log a decision and print the audit trail

Call the log_decision function with input_data set to {"feature1": 5, "feature2": 3} and prediction set to "ClassA". Then print the audit_trail list.

MLOps

audit_trail = []
model_name = "SimpleClassifierV1"

def log_decision(input_data, prediction):
    import datetime
    now = datetime.datetime.now()
    audit_trail.append({
        "model": model_name,
        "input": input_data,
        "prediction": prediction,
        "timestamp": now
    })

# Call log_decision with input_data and prediction
# Print audit_trail
# Your code here

Need a hint?

The printed list should contain one dictionary with the model name, input data, prediction, and a timestamp starting with datetime.datetime.