paulpas Data Privacy
Comprehensive guide to data privacy in machine learning and data science workflows.
When to Use This Skill
- Solving real-world data collection & ingestion problems
- Building machine learning pipelines with data privacy
- Implementing best practices for data privacy
- Optimizing model performance using data privacy techniques
- Learning industry-standard approaches to data privacy
When NOT to Use This Skill
- When using pre-built libraries without understanding underlying concepts
- For toy problems that don't require data privacy rigor
- When domain expertise in specific problem requires different approach
- If your problem doesn't require the complexity this skill provides
Purpose and Key Concepts
Data Privacy is a critical component of the machine learning workflow. This skill covers:
- Theoretical foundations — Mathematical principles and statistical concepts
- Practical implementation — Working code examples and patterns
- Common pitfalls — Mistakes to avoid and how to recover from them
- Best practices — Industry-standard approaches and optimization techniques
Core Workflow
- Understand the problem — Clearly define what you're solving for
- Select approach — Choose the right technique for your data and constraints
- Implement solution — Write clean, tested code following best practices
- Validate results — Verify your implementation with tests and validation
- Optimize performance — Improve efficiency and accuracy incrementally
Implementation Patterns
Pattern 1: Basic Data Privacy
import pandas as pd
import numpy as np
from typing import Dict, Any
TOKEN_SPACE_SIZE: int = 10000
def apply_basic_privacy(data: pd.DataFrame, epsilon: float = 1.0) -> Dict[str, Any]:
"""
Apply basic privacy-preserving transformations.
Implements tokenization for categorical PII and Laplace noise for numerical data.
"""
if data.empty:
raise ValueError("Input DataFrame cannot be empty")
protected_data = data.copy()
metrics: Dict[str, int] = {"columns_anonymized": 0, "noise_added": 0}
for col in protected_data.columns:
if protected_data[col].dtype == "object":
protected_data[col] = protected_data[col].apply(
lambda x: f"token_{abs(hash(str(x))) % TOKEN_SPACE_SIZE}" if pd.notna(x) else x
)
metrics["columns_anonymized"] += 1
elif np.issubdtype(protected_data[col].dtype, np.number):
sensitivity: float = protected_data[col].max() - protected_data[col].min()
scale: float = sensitivity / epsilon
noise: np.ndarray = np.random.laplace(0, scale, size=protected_data[col].shape)
protected_data[col] = protected_data[col] + noise
metrics["noise_added"] += 1
return {"protected_data": protected_data, "metrics": metrics}
Pattern 2: Production-Ready Data Privacy
import logging
import pandas as pd
import numpy as np
from typing import Any, Dict, List
logger = logging.getLogger(__name__)
class DataPrivacyEngine:
"""Production-grade privacy engine implementing GDPR-compliant anonymization."""
def __init__(self, epsilon: float = 1.0, k_anonymity: int = 5) -> None:
self.epsilon: float = epsilon
self.k_anonymity: int = k_anonymity
self._log: logging.Logger = logging.getLogger(self.__class__.__name__)
def execute(self, data: pd.DataFrame, sensitive_cols: List[str]) -> Dict[str, Any]:
"""Execute privacy transformations on sensitive columns."""
if not isinstance(data, pd.DataFrame):
raise TypeError("Expected pandas DataFrame")
if data.empty:
raise ValueError("Input data cannot be empty")
protected: pd.DataFrame = data.copy()
results: Dict[str, Any] = {"status": "success", "transformations_applied": []}
for col in sensitive_cols:
if col not in protected.columns:
self._log.warning(f"Column {col} not found, skipping.")
continue
if protected[col].dtype == "object":
protected[col] = protected[col].apply(
lambda x: f"cat_{abs(hash(str(x))) % 100}" if pd.notna(x) else x
)
results["transformations_applied"].append(f"tokenized_{col}")
else:
col_min: float = protected[col].min()
col_max: float = protected[col].max()
sensitivity: float = col_max - col_min
scale: float = sensitivity / self.epsilon
noise: np.ndarray = np.random.laplace(0, scale, size=len(protected))
protected[col] = protected[col] + noise
results["transformations_applied"].append(f"dp_noised_{col}")
results["protected_data"] = protected
self._log.info(f"Privacy engine completed. Applied {len(results['transformations_applied'])} transformations.")
return results
Best Practices
- ✅ Always validate your implementation on test data
- ✅ Document your assumptions and methodology
- ✅ Use version control for reproducibility
- ✅ Monitor performance metrics in production
- ✅ Periodically review and update your approach
- ✅ Test with edge cases and outliers
- ✅ Log all significant operations for debugging
Common Pitfalls
| Pitfall | Problem | Solution | |
Constraints
MUST DO
- Validate all data preprocessing steps are fit-only on training data, never on validation or test sets
- Implement reproducible pipelines with fixed random seeds and deterministic operations where possible
- Report model performance with confidence intervals via bootstrapping or cross-validation across multiple runs
- Log all experiments with parameters, metrics, and artifacts using MLflow or equivalent tracking system
MUST NOT DO
- Do not evaluate a model on the same data used for training — always hold out a proper test set
- Avoid overfitting to the validation set by limiting hyperparameter search iterations
- Never use features that can only be computed at inference time (look-ahead bias)
- Do not report single-run accuracy without statistical significance testing or error bars
Live References
Authoritative documentation links for this skill's domain. The model follows markdown links at load time to resolve external references and inline content.