Enforce Model Schema Signatures with MLflow

Q: Why do candidates get "Enforce Model Schema Signatures with MLflow" wrong?

Double-counting metrics by using COUNT(*) after joining parent and child tables. When joining an invoice table with an invoice lines table, a single invoice multiplies across all its line items, causing COUNT(invoice_id) to return line counts instead of unique invoice counts.

Q: What are interviewers actually testing on "Enforce Model Schema Signatures with MLflow"?

Interviewers check whether you notice 1-to-many cardinality multiplication and use COUNT(DISTINCT col) or pre-aggregate child records before joining.

Q: What is the optimal query construction strategy for Enforce Model Schema Signatures with MLflow?

Step 1 (Select Core Data Source): Identify the base table and apply preliminary WHERE filters. Step 2 (Compute Group Aggregations): Group by primary business keys and compute aggregate expressions. Step 3 (Order and Limit): Order by specified metrics descending and apply limit clauses.

Question

# Enforce Model Schema Signatures with MLflow

Production machine learning services frequently crash when incoming JSON payloads contain unexpected column names, missing features, or corrupted data types. MLflow solves this via model signatures (`infer_signature`), which bind input feature schemas and output prediction shapes directly into the logged model artifact.

### Task
Write a function `build_model_signature(X_train: pd.DataFrame, model, sample_size: int = 10)` that:
1. Takes a training feature DataFrame `X_train` and a fitted scikit-learn classifier `model`.
2. Generates predictions on a sample slice `X_train.head(sample_size)`.
3. Invokes `mlflow.models.infer_signature` to infer and return the schema `ModelSignature` object.

Accepted Answer

import mlflow.models
import pandas as pd

def build_model_signature(X_train: pd.DataFrame, model, sample_size: int = 10):
    sample_inputs = X_train.head(sample_size)
    sample_predictions = model.predict(sample_inputs)
    return mlflow.models.infer_signature(sample_inputs, sample_predictions)

Candidate Telemetry HUD

Problem Statement

Why 40.8% of Candidates Fail This

What Tech Interviewers Are Testing

Step-by-Step Optimal Query Construction

Select Core Data Source

Compute Group Aggregations

Order and Limit

Common Candidate Wrong Queries & Why They Fail

Double Counting Joined Entities

Core SQL Anti-Patterns to Avoid

Cardinality Inflation from 1:N Joins

COUNT(*) vs COUNT(DISTINCT key)

Integer Division Floor Truncation

Ready to verify your query against real data?

Related SQL Practice Questions

Compute Population Stability Index (PSI) with Epsilon Smoothing

Implement FastAPI Lifespan Model Caching

Deterministic Salted Hash Routing for A/B Canaries

Two-Sample Kolmogorov-Smirnov Drift Test Runner

Longest Substring with At Most K Distinct Characters

Subarray Sum Equals K