milieudefensie 0.0.2

Machine learning tools for environmental analysis

Decision Tree Visualization and Classification Analysis Toolkit

A comprehensive Python package for creating interactive decision tree visualizations and performing advanced classification model analysis.

Features

• Interactive HTML Decision Trees: Visualize decision paths with D3.js
• Automated Model Comparison: Evaluate multiple classifiers with one function
• Feature Importance Analysis: Multiple methods including SHAP, permutation, and native importance
• Probability Analysis: Visualize prediction distributions
• Data Preprocessing: Automatic handling of categorical and numerical variables

Milieudefensie Decision Tree Tools

Python package for creating interactive decision tree visualizations for environmental analysis.

Quick Start Example

Beslisboom

from milieudefensie.beslisboom import CreateHTML
from milieudefensie.model_non_linear_y import generate_example_dataset
import time
import numpy as np

# Generate example data
df = generate_example_dataset(n_samples=10000)
df['Interesse'] = np.where(df['y_binary'] == "yes", 1, 0)

# Configure tree parameters
variables_ = ['major_donor_2024', 'age', 'member']  # Variables to include
y = 'Interesse'  # Target variable for splits
split_method = 'gini'  # 'gini' or 'mean' for splitting criteria
min_records = 500  # Minimum records required for a split
max_integer = 5  # Max splits for integer variables
max_nr_splits = 2  # Max splits except for categoricals
min_split_values = 1000  # Minimum records per split
nr_splits = {'age': 4}  # Custom number of splits per variable
splits = {'age': [20,25,40,60]}  # Custom split points
color_reverse = True  # Color scheme (True: red=low, blue=high)
name_all = 'Selectie interesse nalaten'  # Root node name

# Create and visualize the tree
create_html = CreateHTML(
    df=df,
    vars=variables_,
    y=y,
    split_method=split_method,
    min_records=min_records,
    max_integer=max_integer,
    max_nr_splits=max_nr_splits,
    min_split_values=min_split_values,
    nr_splits=nr_splits,
    splits=splits,
    color_reverse=color_reverse,
    name_all=name_all,
    reorder=False  # Keep original variable order
)

# Generate HTML output
create_html.build_HTML(
    output_file='beslisboom_voorbeeld.html',
    title='Interesse Analysis',
    explanation='This tree shows how different donor characteristics affect legacy interest',
    made_by='Milieudefensie Data Team'
)


# Milieudefensie Machine Learning Toolkit

## Comprehensive Usage Example

```python
from milieudefensie import generate_example_dataset, ClassificationAnalyzer

### 1. Data Preparation
# Generate example dataset (1000 samples)
df = generate_example_dataset(n_samples=1000)

# Select target variable (binary or multiclass)
y = df['y_binary']  # Binary target
# y = df['y_multiclass']  # Multiclass target

# Prepare features
X = df.drop(columns=['y_binary', 'y_multiclass'])

# Generate larger dataset for predictions (10000 samples)
df_all = generate_example_dataset(n_samples=10000)
df_all['relation_id'] = range(1, len(df_all) + 1)
X_all = df_all.drop(columns=['y_binary', 'y_multiclass', 'relation_id'])

### 2. Model Analysis
# Initialize analyzer
analyzer = ClassificationAnalyzer(y=y, X=X)

# Compare multiple models
results = analyzer.compare_models(test_size=0.2)
print("Model Comparison Results:")
print(results['results'])

# Analyze feature importance
feature_importance = analyzer.analyze_feature_importance(method='auto')
print("\nTop Features:")
print(feature_importance.head(10))

### 3. Prediction Generation
# Generate predictions on training data
train_predictions = analyzer.generate_predictions(
    relation_ids=df['age'],  # Using age as ID
    save_as_csv=False
)

# Generate predictions on new data
new_predictions = analyzer.generate_predictions(
    relation_ids=df_all['age'],
    X_new=X_all,
    save_as_csv=True,
    csv_name='new_predictions.csv'
)

### 4. Probability Analysis
# Basic probability analysis
prob_analysis = analyzer.analyze_probabilities()

# Advanced analysis with new data
prob_analysis_new = analyzer.analyze_probabilities(
    X_data=X_all,
    n_cases=1000,
    analysis_type='highest',  # 'highest', 'lowest', or 'both'
    hover_columns=['age', 'gender_male', 'major_donor_2024'],
    plot_filename='interest_probabilities.html'
)

# Multiclass-specific analysis
if len(analyzer.class_names) > 2:
    prob_analysis_custom = analyzer.analyze_probabilities(
        target_class='Interested',  # Specify class to analyze
        n_cases=100,
        analysis_type='highest',
        hover_columns=['age', 'gender_male', 'major_donor_2024'],
        plot_filename='class_probabilities.html'
    )

## Installation

```bash
pip install milieudefensie .