Adult Dataset Example

from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score, roc_auc_score, f1_score, precision_score, recall_score
from sklearn.preprocessing import LabelEncoder
import pandas as pd
import cloudexplain
import datetime

# Load the famous Adult dataset from the UCI repository
github_data_url = "https://github.com/shap/shap/raw/master/data/"
dtypes = [
    ("Age", "float32"),
    ("Workclass", "category"),
    ("fnlwgt", "float32"),
    ("Education", "category"),
    ("Education-Num", "float32"),
    ("Marital Status", "category"),
    ("Occupation", "category"),
    ("Relationship", "category"),
    ("Race", "category"),
    ("Sex", "category"),
    ("Capital Gain", "float32"),
    ("Capital Loss", "float32"),
    ("Hours per week", "float32"),
    ("Country", "category"),
    ("Target", "category"),
]

# Load and preprocess the data
raw_data = pd.read_csv(
    github_data_url + "adult.data", 
    names=[d[0] for d in dtypes], 
    na_values="?", 
    dtype=dict(dtypes)
)

# Clean and prepare the data
data = raw_data.drop(["Education"], axis=1)  # redundant with Education-Num
data["Target"] = data["Target"] == " >50K"  # Convert to binary

# Sample a subset for faster processing
data = data.sample(1000)

# Encode categorical variables
columns_to_encode = [
    'Workclass', 'Marital Status', 'Occupation', 
    'Relationship', 'Race', 'Sex', 'Country'
]

for col in columns_to_encode:
    le = LabelEncoder()
    data[col + '_encoded'] = le.fit_transform(data[col].astype(str))

# Define comprehensive feature descriptions for better explanations
feature_descriptions = {
    'Age': {
        "description": 'Age',
        "display_description": "Age in years",
        "is_categorical": False
    },
    'Workclass_encoded': {
        "description": 'Work Class',
        "display_description": "Type of employment",
        "is_categorical": True,
        "encoding": {
            0: 'Unknown', 1: 'Federal Government', 2: 'Local Government',
            3: 'Never Worked', 4: 'Private Sector', 
            5: 'Self-Employed (Incorporated)', 6: 'Self-Employed (Not Incorporated)',
            7: 'State Government', 8: 'Without Pay'
        }
    },
    'Education-Num': {
        "description": 'Education Level',
        "display_description": "Years of formal education",
        "is_categorical": False
    },
    'Marital Status_encoded': {
        "description": 'Marital Status',
        "display_description": "Current marital status",
        "is_categorical": True,
        "encoding": {
            0: 'Divorced', 1: 'Married (Armed Forces Spouse)',
            2: 'Married (Civilian Spouse)', 3: 'Married (Spouse Absent)',
            4: 'Never Married', 5: 'Separated', 6: 'Widowed'
        }
    },
    'Occupation_encoded': {
        "description": 'Occupation',
        "display_description": "Type of job/profession",
        "is_categorical": True,
        "encoding": {
            0: 'Unknown', 1: 'Administrative/Clerical', 2: 'Armed Forces',
            3: 'Craft/Repair', 4: 'Executive/Managerial', 5: 'Farming/Fishing',
            6: 'Handlers/Cleaners', 7: 'Machine Operator/Inspector',
            8: 'Other Service', 9: 'Private Household Service',
            10: 'Professional Specialty', 11: 'Protective Service',
            12: 'Sales', 13: 'Tech Support', 14: 'Transportation/Moving'
        }
    },
    'Relationship_encoded': {
        "description": 'Family Relationship',
        "display_description": "Relationship within household",
        "is_categorical": True,
        "encoding": {
            0: 'Husband', 1: 'Not in Family', 2: 'Other Relative',
            3: 'Own Child', 4: 'Unmarried Partner', 5: 'Wife'
        }
    },
    'Race_encoded': {
        "description": 'Race',
        "display_description": "Racial background",
        "is_categorical": True,
        "encoding": {
            0: 'American Indian/Eskimo', 1: 'Asian/Pacific Islander',
            2: 'Black', 3: 'Other', 4: 'White'
        }
    },
    'Sex_encoded': {
        "description": 'Gender',
        "display_description": "Gender",
        "is_categorical": True,
        "encoding": {0: 'Female', 1: 'Male'}
    },
    'Capital Gain': {
        "description": 'Capital Gains',
        "display_description": "Capital gains income",
        "is_categorical": False
    },
    'Capital Loss': {
        "description": 'Capital Losses',
        "display_description": "Capital losses",
        "is_categorical": False
    },
    'Hours per week': {
        "description": 'Hours per Week',
        "display_description": "Hours worked per week",
        "is_categorical": False
    },
    'Country_encoded': {
        "description": 'Country',
        "display_description": "Country of origin",
        "is_categorical": True,
        "encoding": {
            0: 'Unknown', 1: 'Cambodia', 2: 'Canada', 3: 'China',
            4: 'Columbia', 5: 'Cuba', 6: 'Dominican Republic', 7: 'Ecuador',
            8: 'El Salvador', 9: 'England', 10: 'France', 11: 'Germany',
            12: 'Greece', 13: 'Guatemala', 14: 'Haiti', 15: 'Netherlands',
            16: 'Honduras', 17: 'Hong Kong', 18: 'Hungary', 19: 'India',
            20: 'Iran', 21: 'Ireland', 22: 'Italy', 23: 'Jamaica',
            24: 'Japan', 25: 'Laos', 26: 'Mexico', 27: 'Nicaragua',
            28: 'US Territories (Guam/USVI)', 29: 'Peru', 30: 'Philippines',
            31: 'Poland', 32: 'Portugal', 33: 'Puerto Rico', 34: 'Scotland',
            35: 'South Korea', 36: 'Taiwan', 37: 'Thailand',
            38: 'Trinidad & Tobago', 39: 'United States', 40: 'Vietnam',
            41: 'Yugoslavia'
        }
    }
}

# Prepare features and target
X = data[['Age', 'Workclass_encoded', 'Education-Num', 'Marital Status_encoded', 
          'Occupation_encoded', 'Relationship_encoded', 'Race_encoded', 'Sex_encoded',
          'Capital Gain', 'Capital Loss', 'Hours per week', 'Country_encoded']]
y = data["Target"]

# Split the data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=42)

# Train the model
model = RandomForestClassifier(random_state=42)
model.fit(X_train, y_train)

# Explain with CloudExplain - This is where the magic happens!
with cloudexplain.azure.explain(
    model=model,
    X=X_test,
    y=y_test,
    model_version="0.1.0",
    model_name="IncomePredictionModel",
    model_description="Predicting whether an individual earns more than $50K annually based on demographic and employment features.",
    explanation_name=f"Income prediction model {datetime.datetime.today().strftime('%Y-%m-%d')}",
    explanation_env="prod",
    data_source="adult dataset",
    resource_group_name="p1-cloudexplain-tf",
    ml_type="binary_classification",
    is_higher_output_better=True,
    feature_descriptions=feature_descriptions,
    baseline_data=X_train,
    api_token="your_api_token_here",  # Get this from /dashboards/analytics/tokens
    function_url="https://your-env-execute-containers.azurewebsites.net/api/upload_via_token"
) as run:
    # Make predictions and evaluate
    y_preds = model.predict(X_test)
    y_proba = model.predict_proba(X_test)[:, 1]

    # Print model performance
    print(f"Explanation ID: {run.run_uuid}")
    print("Model Performance:")
    print(f"Accuracy: {accuracy_score(y_test, y_preds):.3f}")
    print(f"ROC AUC: {roc_auc_score(y_test, y_proba):.3f}")
    print(f"F1 Score: {f1_score(y_test, y_preds):.3f}")
    print(f"Precision: {precision_score(y_test, y_preds):.3f}")
    print(f"Recall: {recall_score(y_test, y_preds):.3f}")
    
    # Your model is now fully explained!
    print("\nView your detailed explanations in the CloudExplain dashboard!")