generate_dataset.py

"""
Dataset Generator: Housing Tenure Security Africa

Parameter Evidence Table:
| Parameter | Value | Source |
|-----------|-------|--------|
| Informal tenure prevalence | 60-80% in slums | UN-Habitat 2023 |
| Rental housing share | 30-50% urban | OECD Africa Urbanization 2025 |
| Title deed availability | 10-30% informal | World Bank land data |
| Eviction risk in informal | 15-40% | UN-Habitat estimates |
| Year range | 2018-2025 | Project scope |
| Countries | 15 African nations | Regional coverage |

DAG Structure:
country -> tenure_type -> security_level -> documentation -> risk_factors
"""

import numpy as np
import pandas as pd
from pathlib import Path

COUNTRIES = [
    'Nigeria', 'Kenya', 'Ethiopia', 'Egypt', 'South Africa',
    'Tanzania', 'Morocco', 'Algeria', 'Ghana', 'Uganda',
    'Mozambique', 'Zambia', 'Malawi', 'Rwanda', 'Senegal'
]

SCENARIOS = {
    'low_burden': {'n': 4000, 'informal_pct': 0.45, 'seed': 42},
    'moderate': {'n': 5000, 'informal_pct': 0.60, 'seed': 43},
    'high': {'n': 6000, 'informal_pct': 0.75, 'seed': 44}
}

CITIES = {
    'Nigeria': ['Lagos', 'Kano', 'Ibadan'],
    'Kenya': ['Nairobi', 'Mombasa', 'Kisumu'],
    'Ethiopia': ['Addis Ababa', 'Dire Dawa'],
    'Egypt': ['Cairo', 'Alexandria'],
    'South Africa': ['Johannesburg', 'Cape Town'],
    'Tanzania': ['Dar es Salaam'],
    'Morocco': ['Casablanca', 'Rabat'],
    'Algeria': ['Algiers'],
    'Ghana': ['Accra', 'Kumasi'],
    'Uganda': ['Kampala'],
    'Mozambique': ['Maputo', 'Beira'],
    'Zambia': ['Lusaka'],
    'Malawi': ['Lilongwe', 'Blantyre'],
    'Rwanda': ['Kigali'],
    'Senegal': ['Dakar']
}

def dag_sample_tenure_type(informal_pct, rng):
    weights = [informal_pct, 0.35, 0.15, 0.08]
    weights = [w / sum(weights) for w in weights]
    return rng.choice(['informal', 'rental', 'owner_occupied', 'customary'], p=weights)

def dag_sample_security_level(tenure_type, informal_pct, rng):
    security_map = {
        'informal': rng.uniform(0.2, 0.5),
        'rental': rng.uniform(0.4, 0.7),
        'owner_occupied': rng.uniform(0.7, 0.95),
        'customary': rng.uniform(0.3, 0.6)
    }
    if tenure_type == 'informal':
        security_map['informal'] *= (1 - informal_pct * 0.3)
    return round(security_map[tenure_type], 3)

def dag_sample_documentation(tenure_type, security, rng):
    doc_chances = {
        'informal': 0.15,
        'rental': 0.55,
        'owner_occupied': 0.75,
        'customary': 0.25
    }
    has_title = rng.random() < doc_chances.get(tenure_type, 0.3)
    
    return {
        'has_title_deed': has_title,
        'has_formal_lease': rng.random() < doc_chances.get(tenure_type, 0.3) * 0.8,
        'proof_of_occupancy': rng.random() < 0.7,
        'registration_status': rng.choice(['registered', 'pending', 'none'])
    }

def generate_dataset(scenario, output_dir):
    config = SCENARIOS[scenario]
    rng = np.random.default_rng(config['seed'])
    
    data = {
        'country': [],
        'city': [],
        'year': [],
        'tenure_type': [],
        'security_score': [],
        'has_title_deed': [],
        'has_formal_lease': [],
        'proof_of_occupancy': [],
        'registration_status': [],
        'years_in_tenure': [],
        'eviction_risk_pct': [],
        'perceived_security': [],
        'willingness_to_invest': [],
        'tenure_type_detail': [],
        'landlord_relationship': [],
        'rent_control_applicable': []
    }
    
    for _ in range(config['n']):
        country = rng.choice(COUNTRIES)
        city = rng.choice(CITIES.get(country, ['Unknown']))
        year = rng.integers(2018, 2026)
        
        tenure_type = dag_sample_tenure_type(config['informal_pct'], rng)
        security = dag_sample_security_level(tenure_type, config['informal_pct'], rng)
        docs = dag_sample_documentation(tenure_type, security, rng)
        
        eviction_risk = (1 - security) * 100
        
        data['country'].append(country)
        data['city'].append(city)
        data['year'].append(year)
        data['tenure_type'].append(tenure_type)
        data['security_score'].append(security)
        data['has_title_deed'].append(docs['has_title_deed'])
        data['has_formal_lease'].append(docs['has_formal_lease'])
        data['proof_of_occupancy'].append(docs['proof_of_occupancy'])
        data['registration_status'].append(docs['registration_status'])
        data['years_in_tenure'].append(int(rng.uniform(1, 25)))
        data['eviction_risk_pct'].append(round(eviction_risk, 1))
        data['perceived_security'].append(rng.choice(['very_secure', 'secure', 'insecure', 'very_insecure']))
        data['willingness_to_invest'].append(round(rng.uniform(0.1, 0.9), 2))
        data['tenure_type_detail'].append(rng.choice(['formal', 'informal', 'customary', 'communal']))
        data['landlord_relationship'].append(rng.choice(['family', 'private', 'government', 'community']))
        data['rent_control_applicable'].append(rng.choice([True, False], p=[0.2, 0.8]))
    
    df = pd.DataFrame(data)
    output_dir.mkdir(parents=True, exist_ok=True)
    df.to_csv(output_dir / f'{scenario}.csv', index=False)
    print(f"Generated {scenario}: {len(df)} rows")

if __name__ == '__main__':
    base_dir = Path(__file__).parent
    for scenario in SCENARIOS:
        generate_dataset(scenario, base_dir)