Evaluate the Responsible AI dashboard and see what it does.

Here's a structured list of Responsible AI Dashboard Deployment Steps using the corrected scripts. Each step includes:

• ✅ Step Number & Action
• 🎯 Purpose
• 🧪 Expected Result

pip install --upgrade raiutils raiwidgets responsibleai ipywidgets

🎯 Purpose:
Install the Python packages required to run Responsible AI analysis and render the dashboard.

🧪 Expected Result:
Packages are installed without errors; dashboard widgets can render in the notebook (after kernel restart).

from sklearn.datasets import fetch_openml
import pandas as pd

data = fetch_openml(name='adult', version=2, as_frame=True)
df = data.frame.dropna()

🎯 Purpose:
Load a well-known classification dataset (Adult Census Income) and remove any missing values to avoid downstream errors.

🧪 Expected Result:
A clean DataFrame with no null values is loaded.

from sklearn.model_selection import train_test_split

target_column = 'class'
X = df.drop(columns=[target_column])
y = df[target_column]

X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, random_state=42)

🎯 Purpose:
Separate features and target, then split into training/testing sets for model training and evaluation.

🧪 Expected Result:
X_train, X_test, y_train, y_test variables created and stratified properly.

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.ensemble import RandomForestClassifier

categorical_cols = X_train.select_dtypes(include=['object', 'category']).columns.tolist()
numerical_cols = X_train.select_dtypes(include=['int64', 'float64']).columns.tolist()

preprocessor = ColumnTransformer([
    ('cat', OneHotEncoder(handle_unknown='ignore'), categorical_cols),
    ('num', StandardScaler(), numerical_cols)
])

clf = Pipeline(steps=[
    ('preprocessor', preprocessor),
    ('classifier', RandomForestClassifier(n_estimators=10, random_state=42))
])

clf.fit(X_train, y_train)

🎯 Purpose:
Build a model pipeline that encodes categorical features, scales numeric ones, and trains a classifier.

🧪 Expected Result:
Pipeline is trained successfully on the training data without conversion errors.

# Ensure target column is a supported type
y_train_clean = y_train.astype(str)
y_test_clean = y_test.astype(str)

train_data = X_train.copy()
train_data[target_column] = y_train_clean

test_data = X_test.copy()
test_data[target_column] = y_test_clean

🎯 Purpose:
Re-attach the target column (as string) to the feature DataFrames — required for RAIInsights.

🧪 Expected Result:
train_data and test_data DataFrames contain all required columns including the target.

from responsibleai import RAIInsights, FeatureMetadata

feature_metadata = FeatureMetadata(categorical_features=categorical_cols)

rai_insights = RAIInsights(
    model=clf,
    train=train_data,
    test=test_data,
    target_column=target_column,
    task_type="classification",
    feature_metadata=feature_metadata
)

🎯 Purpose:
Create a RAIInsights object that acts as the core engine for the Responsible AI dashboard.

🧪 Expected Result:
RAIInsights object is initialized successfully and ready for configuration.

rai_insights.explainer.add()
rai_insights.error_analysis.add()
rai_insights.counterfactual.add(total_CFs=5, desired_class='opposite')
rai_insights.causal.add(treatment_features=categorical_cols)

🎯 Purpose:
Attach various tools (explanation, error analysis, counterfactuals, causal inference) to the insights engine.

🧪 Expected Result:
No errors thrown; tools are queued for computation.

rai_insights.compute()

🎯 Purpose:
Run analysis for all selected tools. This step may take a minute or more.

🧪 Expected Result:
Tool outputs are generated for the first 5,000 rows of the test set.

from raiwidgets import ResponsibleAIDashboard

ResponsibleAIDashboard(rai_insights)

🎯 Purpose:
Open an interactive dashboard to explore insights such as feature importance, what-if analysis, and error breakdowns.

🧪 Expected Result:
A dashboard is displayed inside the notebook. Interactive plots and controls are available for analysis.

Download

NOTE:  Due to the way that these URLs are deployed, this step will fail because the notebook sends the wrong headers and this is expected.  You have to either pull the notebook local and use it from the terminal or register the dashboard/dataset and review it through the portal.

AI Knowledge