### Manual Audit Implementation with Venturalitica

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/quickstart.md

Demonstrates the manual steps equivalent to the quickstart function, including loading data from UCI repository, defining policies, and running the audit using vl.enforce. This shows the underlying components of the quickstart process.

```python
from ucimlrepo import fetch_ucirepo
import venturalitica as vl

# 1. Load Data (The "Risk Source")
dataset = fetch_ucirepo(id=144)
df = dataset.data.features
df['class'] = dataset.data.targets

# 2. Define the Policy (The "Law")
# We load a pre-defined policies/risks.oscal.yaml

# 3. Run the Audit (The "Test")
# This automatically generates the Evidence Bill of Materials (BOM)
with vl.monitor("manual_audit"):
    vl.enforce(
        data=df,
        target="class",          # The outcome (True/False)
        gender="Attribute9",     # Protected Group A
        age="Attribute13",       # Protected Group B
        policy="risks.oscal.yaml"
    )
```

--------------------------------

### Install Venturalitica SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/quickstart.mdx

Installs the Venturalitica SDK using pip. Optional extras can be installed for additional capabilities like fairness metrics, dashboard, agent support, energy tracking, or a full installation.

```bash
pip install venturalitica

```

--------------------------------

### Run First Audit with Venturalitica

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/quickstart.md

Executes the quickstart function from the Venturalitica SDK to perform an initial audit on the 'loan' scenario. This function automates data loading, policy enforcement, and evidence recording.

```python
import venturalitica as vl

vl.quickstart('loan')
```

--------------------------------

### Quickstart: Run Bias Audit on UCI Dataset

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/README.md

A simple Python script to quickly start using the Venturalítica SDK. It automatically downloads the UCI German Credit dataset and runs a bias audit, providing immediate feedback on fairness.

```python
import venturalitica as vl

# Auto-downloads UCI German Credit and runs bias audit
results = vl.quickstart('loan')
```

--------------------------------

### Import Venturalítica and Check Version (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/00-quickstart.ipynb

Imports the Venturalítica library and prints its version. This is a standard first step to ensure the SDK is installed and accessible.

```python
import venturalitica as vl
print(f"Venturalítica version: {vl.__version__}")
```

--------------------------------

### Quickstart Venturalitica SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/full-lifecycle.mdx

Performs a quick smoke test of the Venturalitica SDK by running a 'loan' scenario. It checks for expected results, indicating a basic setup is functional.

```python
import venturalitica as vl
results = vl.quickstart("loan")
# Expected: 2 PASS, 1 FAIL (age disparity)
```

--------------------------------

### Run Quickstart Audit with Loan Policy (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/00-quickstart.ipynb

Executes a pre-defined fairness audit for loan applications using the `quickstart` function. This function automatically fetches the German Credit dataset, loads a loan fairness policy, and generates a report.

```python
vl.quickstart('loan')
```

--------------------------------

### Run Quickstart Demo with Venturalitica SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/reference/api.mdx

The `quickstart()` function provides a fast way to demonstrate the Venturalitica SDK's capabilities by running a pre-configured bias audit on a standard dataset. It accepts a scenario name and an optional verbose flag to control console output. Internally, it sets up a dataset, loads an OSCAL policy, and calls `enforce()`.

```python
import venturalitica as vl

results = vl.quickstart("loan")
```

--------------------------------

### Run Quickstart Bias Audit

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/quickstart.mdx

Executes a pre-defined bias audit using the Venturalitica SDK's quickstart function. This function automatically loads a dataset and a built-in policy to perform the audit.

```python
import venturalitica as vl

vl.quickstart('loan')

```

--------------------------------

### Run Quickstart Bias Audit (Python)

Source: https://context7.com/venturalitica/venturalitica-sdk/llms.txt

Executes a 60-second bias audit demo using the `quickstart` function. It automatically downloads a pre-configured dataset (UCI German Credit) and uses an embedded OSCAL policy for fairness checks, providing a summary of compliance results.

```python
import venturalitica as vl

# Run pre-configured loan credit scoring bias audit
# Auto-downloads UCI German Credit dataset (1000 samples)
# Uses embedded OSCAL policy for fairness checks
results = vl.quickstart('loan')

# Output:
# [Venturalítica] 🎓 Scenario: Credit Scoring Fairness
# [Venturalítica] 📊 Loaded: UCI Dataset #144 (1000 samples)
# [Venturalítica] 🛡️  Policy: Embedded (no file required)
#
#   CONTROL                DESCRIPTION                            ACTUAL     LIMIT      RESULT
#   ────────────────────────────────────────────────────────────────────────────────────────────────
#   credit-data-imbalance  Data Quality                           0.431      > 0.2      ✅ PASS
#   credit-data-bias       Disparate impact                       0.836      > 0.8      ✅ PASS
#   credit-age-disparate   Age disparity                          0.361      > 0.5      ❌ FAIL
#   ────────────────────────────────────────────────────────────────────────────────────────────────
#   Audit Summary: ❌ VIOLATION | 2/3 controls passed
#
# 🎉 Aha! Moment: Age bias detected in UCI German Credit dataset!

# Returns List[ComplianceResult] for programmatic access
for result in results:
    print(f"{result.control_id}: {result.passed} (value: {result.actual_value})")
```

--------------------------------

### Install Venturalitica SDK using uv or pip

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/academy/index.md

Installs the Venturalitica SDK from its GitHub repository. It supports installation using either 'uv' for faster performance or the standard 'pip' package manager.

```bash
uv pip install git+https://github.com/Venturalitica/venturalitica-sdk.git
```

```bash
pip install git+https://github.com/Venturalitica/venturalitica-sdk.git
```

--------------------------------

### Install Venturalitica SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/full-lifecycle.mdx

Installs the Venturalitica SDK using pip. Optional extras like fairness metrics can be installed with `pip install "venturalitica[metrics]"`.

```bash
pip install venturalitica
```

--------------------------------

### Install Venturalítica SDK (Bash)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/index.md

Provides the command to install the Venturalítica SDK directly from its GitHub repository using pip. This ensures you have the latest version for local deployment.

```bash
pip install git+https://github.com/Venturalitica/venturalitica-sdk.git
```

--------------------------------

### List Available Scenarios

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/api.md

Retrieves a dictionary containing all available pre-configured scenarios and their corresponding descriptions. This utility function helps users understand the options for the `quickstart` function.

```python
import venturalitica as vl

available_scenarios = vl.list_scenarios()
print(available_scenarios)
```

--------------------------------

### Manual Bias Audit Workflow

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/quickstart.mdx

Demonstrates the manual steps involved in running a bias audit, including data loading, policy definition (using an OSCAL file), and enforcing the audit with specified attributes and a target.

```python
from ucimlrepo import fetch_ucirepo
import venturalitica as vl

# 1. Load Data (The "Risk Source")
dataset = fetch_ucirepo(id=144)
df = dataset.data.features
df['class'] = dataset.data.targets

# 2. Define the Policy (The "Law")
# quickstart() uses a built-in policy dict.
# In real projects, you write your own OSCAL YAML file.
# See the Full Lifecycle guide for a copy-paste example.

# 3. Run the Audit (The "Test")
# This automatically generates the Evidence Bill of Materials (BOM)
with vl.monitor("manual_audit"):
    vl.enforce(
        data=df,
        target="class",          # The outcome (True/False)
        gender="Attribute9",     # Protected Group A
        age="Attribute13",       # Protected Group B
        policy="data_policy.oscal.yaml"  # Your OSCAL policy file
    )

```

--------------------------------

### GitHub Actions Workflow for Venturalitica Compliance

Source: https://context7.com/venturalitica/venturalitica-sdk/llms.txt

Example GitHub Actions workflow demonstrating how to set up a compliance gate using Venturalitica's Strict Mode. The workflow installs the SDK and runs a Python script to enforce compliance policies, failing the build if any control fails or configuration is missing.

```yaml
name: Compliance Gateon: [push, pull_request]

jobs:
  compliance:
    runs-on: ubuntu-latest
    env:
      VENTURALITICA_STRICT: true
    steps:
      - uses: actions/checkout@v3
      - name: Run Compliance Audit
        run: |
          pip install venturalitica
          python -c "
          import venturalitica as vl
          import pandas as pd
          df = pd.read_csv('data/test.csv')
          vl.enforce(data=df, target='label', policy='policy.yaml')
          "
      # Build fails if any control fails or config is missing
```

--------------------------------

### Core Functions - quickstart

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/api.md

Runs a pre-configured bias audit demo on a standard dataset. It accepts a scenario name and an optional verbose flag.

```APIDOC
## POST /quickstart

### Description
Run a pre-configured bias audit demo on a standard dataset.

### Method
POST

### Endpoint
/quickstart

### Parameters
#### Query Parameters
- **scenario** (str) - Required - Predefined scenario: `'loan'`, `'hiring'`, `'health'`.
- **verbose** (bool) - Optional - Whether to print the structured table report to the console.

### Response
#### Success Response (200)
- **results** (List[ComplianceResult]) - A list of compliance results from the audit.

### Request Example
```json
{
  "scenario": "loan",
  "verbose": true
}
```

### Response Example
```json
{
  "results": [
    {
      "metric": "Demographic Parity",
      "value": 0.85,
      "threshold": 0.8
    }
  ]
}
```
```

--------------------------------

### Quickstart Audit with Venturalítica SDK (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/index.md

Demonstrates how to perform a quick audit using the Venturalítica SDK with a single line of Python code. This function automatically records evidence for compliance.

```python
import venturalitica as vl

# 1. Run Audit (Auto-Records Evidence)
results = vl.quickstart('loan')
```

--------------------------------

### Install Venturalitica SDK using pip (Bash)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/RELEASE_NOTES.md

Command to install a specific version of the Venturalitica SDK using pip. This is useful for users who need to ensure they have version 0.5.0 installed.

```bash
pip install venturalitica-sdk==0.5.0
```

--------------------------------

### Launch Venturalítica UI

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/annex-iv.md

Starts the Venturalítica Dashboard from the project's root directory. This UI allows for interactive generation of Annex IV documentation.

```bash
venturalitica ui
```

--------------------------------

### Clone Venturalitica SDK Samples Repository

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/academy/index.md

Clones the Venturalitica SDK samples repository to your local machine. This repository serves as the working directory for various levels of AI governance training and practical examples.

```bash
git clone https://github.com/venturalitica/venturalitica-sdk-samples.git
cd venturalitica-sdk-samples/scenarios/loan-credit-scoring
```

--------------------------------

### Run Quickstart Bias Audit Demo

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/api.md

Executes a pre-configured bias audit demo using a standard dataset. It requires a scenario string and an optional verbose boolean for console output. The function returns a list of ComplianceResult objects.

```python
import venturalitica as vl

# Example usage:
results = vl.quickstart(scenario='loan', verbose=True)
```

--------------------------------

### Import Libraries and Check Venturalítica Version (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/01-training-tutorial.ipynb

Imports necessary libraries for data manipulation, ML modeling, and Venturalítica SDK. It also prints the installed Venturalítica version.

```python
from ucimlrepo import fetch_ucirepo
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
import venturalitica as vl
import pandas as pd
import yaml

print("Venturalítica version:", vl.__version__ if hasattr(vl, '__version__') else '0.2.0')
```

--------------------------------

### Install Optional Dependencies for Probes

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/reference/probes.mdx

Some Venturalitica SDK probes have optional dependencies that enhance their functionality. The HardwareProbe can utilize `psutil` for detailed CPU and memory usage, while the CarbonProbe requires `codecarbon` for tracking energy consumption and carbon emissions. These can be installed using pip.

```bash
pip install psutil codecarbon
```

--------------------------------

### Artifact Tracking with Input/Output Paths (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/reference/probes.mdx

Illustrates how to use the ArtifactProbe by specifying input and output file paths during monitor initialization. This captures snapshots of artifacts at the start and end of the monitored execution.

```python
with vl.monitor("training",
                inputs=["data/train.csv"],
                outputs=["models/credit_model.pkl"]):
    model.fit(X, y)
```

--------------------------------

### Install mdpdf for Document Conversion

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/full-lifecycle.mdx

Installs the 'mdpdf' Python package, which is used to convert Markdown files to PDF format. This is typically used after generating regulatory documents like Annex IV.

```bash
pip install mdpdf
```

--------------------------------

### Install Venturalitica SDK using uv (Bash)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/RELEASE_NOTES.md

Command to add a specific version of the Venturalitica SDK to a project using the uv package installer. This is an alternative to pip for managing Python dependencies.

```bash
uv add venturalitica-sdk==0.5.0
```

--------------------------------

### Run Venturalitica UI

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/academy/level2_integrator.md

This command starts the Venturalitica UI, allowing for verification of shipped code and results. It is a prerequisite for checking policy status and risk treatment.

```bash
uv run venturalitica ui
```

--------------------------------

### Group Controls in YAML Policy

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/guides/policy-authoring.mdx

This YAML example demonstrates how to group multiple controls within a single policy file under logical descriptions. It shows sections for 'Data Quality Controls', 'Fairness Controls', and 'Privacy Controls', allowing for organized policy definitions.

```yaml
assessment-plan:
  metadata:
    title: "Comprehensive AI Assurance Policy"
  control-implementations:
    - description: "Data Quality Controls (Article 10)"
      implemented-requirements:
        - control-id: dq-001
          # ...
        - control-id: dq-002
          # ...

    - description: "Fairness Controls (Article 9)"
      implemented-requirements:
        - control-id: fair-001
          # ...

    - description: "Privacy Controls (GDPR)"
      implemented-requirements:
        - control-id: priv-001
          # ...
```

--------------------------------

### Convert Markdown to PDF using mdpdf

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/academy/level4_annex_iv.md

This command installs the mdpdf Python package and then uses it to convert a Markdown file (Annex_IV.md) into a PDF document. Ensure you have Python and pip/uv installed.

```bash
uv pip install mdpdf
uv run mdpdf Annex_IV.md
```

--------------------------------

### Educational Audit Control Description Example

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/README.md

An example of a control definition within a policy file that includes an educational description. This description explains the rationale behind the data quality check, specifically regarding class imbalance.

```yaml
- control-id: data-quality-check
  description: "Data Quality: Minority class should represent at least 20% to avoid Class Imbalance"
```

--------------------------------

### Run First AI Audit with Venturalitica SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/academy/index.md

Executes a quickstart function from the Venturalitica SDK to perform an initial audit on a loan credit scoring model. This function automatically downloads a dataset, loads a predefined policy, and audits the model, providing a PASS/FAIL result for each control.

```python
import venturalitica as vl

# Run the 'loan' scenario
vl.quickstart('loan')
```

--------------------------------

### Install Venturalítica SDK Packages

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/index.mdx

Installs the Venturalítica SDK with different feature sets. The core package is required, while others like 'metrics', 'dashboard', 'agentic', or 'full' can be added for extended functionality. These commands are executed using pip.

```bash
pip install venturalitica
pip install "venturalitica[metrics]"
pip install "venturalitica[dashboard]"
pip install "venturalitica[agentic]"
pip install "venturalitica[full]"
```

--------------------------------

### Load UCI German Credit Dataset (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/01-training-tutorial.ipynb

Loads the UCI German Credit dataset using `fetch_ucirepo` and splits it into training and testing sets. It prints the total number of samples loaded.

```python
dataset = fetch_ucirepo(id=144)
df = dataset.data.features.copy()
df['class'] = dataset.data.targets

train_df, test_df = train_test_split(df, test_size=0.2, random_state=42)
print(f"Loaded {len(df)} samples.")
df.head()
```

--------------------------------

### Enforce Multiple Policies with Python SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/guides/policy-authoring.mdx

Illustrates enforcing multiple policies (data and model) in a single call using the Venturalitica Python SDK. It explains how controls are skipped if required inputs are missing and shows how to supply all necessary inputs for both policy types to be evaluated.

```python
# Only data-policy controls will run here because
# 'prediction' is not provided — model-policy controls are skipped.
vl.enforce(
    data=df,
    target="class",
    gender="Attribute9",
    policy=["data_policy.oscal.yaml", "model_policy.oscal.yaml"]
)

# To evaluate both policies, supply all required inputs:
vl.enforce(
    data=test_df,
    target="class",
    prediction="y_pred",
    gender="Attribute9",
    policy=["data_policy.oscal.yaml", "model_policy.oscal.yaml"]
)
```

--------------------------------

### Monitor Model Training with vl.monitor

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/01-training-tutorial.ipynb

This code snippet shows how to use `vl.monitor` to track metadata during model training. It wraps the `model.fit` method within a `vl.monitor` context, allowing for the collection of metrics such as duration, hardware usage, and carbon emissions. The monitor is started with a specific name and stopped automatically when the context manager exits.

```python
model = RandomForestClassifier(n_estimators=100, random_state=42)

with vl.monitor(name="RandomForest-Credit"):
    model.fit(X_train, y_train)
    
predictions = model.predict(X_test)
```

--------------------------------

### Monitor Model Training with Venturalitica Python

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/reference/api.mdx

This snippet demonstrates how to use the `vl.monitor` context manager to automatically collect probes during model training. It initializes 7 probes that capture various aspects of the training environment and process, such as hardware usage, CO2 emissions, software bill of materials, data lineage, and code analysis. Evidence is saved to a local directory.

```python
import venturalitica as vl

with vl.monitor("credit_model_v1"):
    model.fit(X_train, y_train)
    vl.enforce(data=df, policy="policy.oscal.yaml", target="class")
```

--------------------------------

### Verify Data Policy with Venturalitica SDK (Python)

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/tutorials/01_writing_policy.md

Demonstrates how to use the Venturalitica SDK in Python to load sample data and enforce a data policy defined in a YAML file.

```python
import venturalitica as vl
from venturalitica.quickstart import load_sample

# 1. Load the 'Approved' Dataset (Mock)
data = load_sample('loan')

# 2. Dry Run the Policy
try:
    vl.enforce(
        data=data,
        target="class",
        gender="Attribute9",  # "Personal status and sex" in German Credit Data
        policy="data_policy.yaml"
    )
    print("✅ Policy is valid syntax and passes baseline data.")
except Exception as e:
    print(f"❌ Policy Error: {e}")
```

--------------------------------

### Train Random Forest Model with Python

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/training.md

Trains a RandomForestClassifier model using scikit-learn on the pre-processed training data. It then generates predictions on the test set for subsequent model auditing.

```python
from sklearn.ensemble import RandomForestClassifier

print("🤖 Training Model...")
model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)

# Generate predictions for the next audit
predictions = model.predict(X_test)
```

--------------------------------

### Enforce Data and Model Policies using Python SDK

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/starlight/src/content/docs/guides/policy-authoring.mdx

Demonstrates how to use the Venturalitica Python SDK to enforce data and model policies. It shows separate calls for pre-training data audits and post-training model audits, highlighting the `enforce` function with different dataframes and policy files.

```python
# Pre-training: audit the data
vl.enforce(data=train_df, target="class", gender="Attribute9",
           policy="data_policy.oscal.yaml")

# Post-training: audit the model
vl.enforce(data=test_df, target="class", prediction="y_pred",
           gender="Attribute9", policy="model_policy.oscal.yaml")
```

--------------------------------

### Define Fairness Policy for Auditing

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/README.md

An example YAML file defining a fairness policy for the Venturalítica SDK. This policy specifies checks for gender-based approval rates using the demographic parity difference metric.

```yaml
assessment-plan:
  uuid: my-policy
  metadata:
    title: "Fairness Policy"
  reviewed-controls:
    control-selections:
      - include-controls:
        - control-id: gender-check
          description: "Approval rates must be similar across genders"
          props:
            - name: metric_key
              value: demographic_parity_diff
            - name: threshold
              value: "0.10"
            - name: operator
              value: "<"
```

--------------------------------

### Run Loan Scenario with Alternative Metrics

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/METRICS_GUIDE.md

Executes a Python script demonstrating the use of alternative fairness metrics within a loan scenario. This is useful for understanding how different fairness definitions apply in a practical context.

```bash
cd scenarios/loan-mlflow-sklearn
python 02_alternative_metrics_demo.py
```

--------------------------------

### Load Sample Dataset by Scenario

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/docs/api.md

Loads a sample dataset from the UCI repository corresponding to a specified scenario. The loaded data is returned as a Pandas DataFrame, suitable for use with other Venturalitica functions like `enforce` or `quickstart`.

```python
import venturalitica as vl

# Example: Load the 'loan' scenario dataset
dataframe = vl.load_sample(scenario='loan')
print(dataframe.head())
```

--------------------------------

### Automatic Governance with vl.wrap - Python

Source: https://github.com/venturalitica/venturalitica-sdk/blob/main/notebooks/01-training-tutorial.ipynb

This code snippet shows how to automatically execute audits using `vl.wrap`. It preprocesses the data, wraps a Scikit-Learn model, and then calls `.fit()` with the original training data and attribute mappings. This automates the audit process every time the model is trained or used for prediction.

```python
# Pre-processing as before
df_encoded = pd.get_dummies(df.drop(columns=['class']))
X_train, X_test, y_train, y_test = train_test_split(
    df_encoded, 
    df['class'].values.ravel(), 
    test_size=0.2, 
    random_state=42
)

# Wrap your model
base_model = RandomForestClassifier(n_estimators=100, random_state=42)
governed_model = vl.wrap(base_model, policy="policy.oscal.yaml")

# Audits automated! Just provide the raw data for attribution mapping
governed_model.fit(X_train, y_train, audit_data=train_df, gender="Attribute9", age="Attribute13")
```