pyrevealed 0.4.0

Behavioral Signal Analysis for User Understanding - Detect bots, shared accounts, and UI confusion

PyRevealed

A Python implementation of revealed preference theory.

Based on: Chambers, C. P., & Echenique, F. (2016). Revealed Preference Theory. Cambridge University Press.

What is this?

Given a history of user choices and the options available at each choice, PyRevealed computes:

• Consistency scores: How internally consistent is this user's behavior? (0 = random, 1 = perfectly consistent)
• Preference recovery: If consistent, what utility function explains their choices?
• Exploitability metrics: How much could be extracted from a user via arbitrage on their inconsistencies?
• Feature independence: Are choices over group A independent of choices over group B?

Installation

pip install pyrevealed

For visualization support:

pip install pyrevealed[viz]

Quick Start

from pyrevealed import BehaviorLog, validate_consistency, compute_integrity_score, compute_confusion_metric
import numpy as np

# Create a behavior log from observed choices
log = BehaviorLog(
    cost_vectors=np.array([      # Prices at each observation (T x N)
        [1.0, 2.0],              # Observation 0: price of good A=1, B=2
        [2.0, 1.0],              # Observation 1: price of good A=2, B=1
    ]),
    action_vectors=np.array([    # Quantities chosen (T x N)
        [3.0, 1.0],              # Observation 0: bought 3 of A, 1 of B
        [1.0, 3.0],              # Observation 1: bought 1 of A, 3 of B
    ])
)

# Test consistency (GARP)
is_consistent = validate_consistency(log)
print(f"Consistent: {is_consistent}")

# Compute integrity score (Afriat Efficiency Index)
integrity = compute_integrity_score(log)
print(f"Integrity Score: {integrity:.3f}")

# Compute confusion metric (Money Pump Index)
confusion = compute_confusion_metric(log)
print(f"Confusion Metric: {confusion:.3f}")

---

Available Tests & Scores

Yes/No Tests

Method	Question it answers
validate_consistency(log)	Is this user rational? (no self-contradicting choices)
validate_consistency_weak(log)	Any obvious flip-flops? (picked A over B, then B over A)
validate_smooth_preferences(log)	Smooth preferences? (needed for price sensitivity analysis)
validate_strict_consistency(log)	Approximately rational? (ignores minor contradictions)
validate_price_preferences(log)	Does user prefer situations where their items are cheaper?

Scores (0 to 1)

Method	What it measures
compute_integrity_score(log)	How consistent is this user? (higher = more consistent)
compute_confusion_metric(log)	How exploitable via pricing tricks? (lower = safer)
compute_minimal_outlier_fraction(log)	Fraction of observations to remove for consistency
compute_test_power(log)	Statistical power of consistency test

Preference Structure

Method	Question it answers
validate_proportional_scaling(log)	Do they buy the same mix regardless of budget size?
test_income_invariance(log)	Does budget size affect what they choose?
test_feature_independence(log, [a], [b])	Are choices in group A separate from group B?
test_cross_price_effect(log, item1, item2)	Are these items substitutes or complements?
transform_to_characteristics(log, A)	Analyze by attributes (nutrition, specs) not products

---

Case Study

See DUNNHUMBY.md for a real-world validation on 2,222 households from the Dunnhumby grocery dataset.

Key findings: 4.5% fully consistent, mean integrity 0.839, test power 0.845.

---

Project Structure

pyrevealed/
├── src/pyrevealed/
│   ├── auditor.py       # BehavioralAuditor class
│   ├── encoder.py       # PreferenceEncoder class
│   ├── lancaster.py     # Lancaster characteristics model
│   ├── algorithms/      # Core algorithms
│   ├── core/            # Data containers
│   ├── graph/           # Graph algorithms
│   └── viz/             # Visualization
├── tests/               # Unit tests
├── dunnhumby/           # Real-world validation suite
│   ├── run_all.py       # Main test runner
│   ├── extended_analysis.py  # Statistical analyses
│   ├── comprehensive_analysis.py  # MPI, WARP, separability
│   ├── advanced_analysis.py  # Complementarity, stress tests
│   ├── encoder_analysis.py  # Auto-discovery, Houtman-Maks
│   ├── predictive_analysis.py  # Split-sample LightGBM
│   ├── lancaster_analysis.py  # Lancaster characteristics model
│   └── data/            # Kaggle dataset (download required)
├── docs/images/         # README visualizations
├── notebooks/           # Tutorials
└── examples/            # Advanced usage examples

License

MIT