Quickstart

In this quickstart, you’ll build a complete ML pipeline that:

1. Loads the Titanic dataset
2. Splits into train/test sets
3. Trains a Random Forest classifier
4. Evaluates accuracy and generates explanations

Time to complete: ~5 minutes

Prerequisites

Before starting, make sure you have:

• MLOps Desktop installed
• Python 3.9+ with packages: pip install scikit-learn pandas shap

Create a Pipeline

1. Open MLOps Desktop

   Launch the app. You’ll see an empty canvas with a toolbar at the top and a node palette on the left.

2. Add a DataLoader node

   From the Components panel on the left, drag DataLoader onto the canvas.

   Click the node to select it. In the node, click Browse and select a CSV file. For this tutorial, use any classification dataset (or see “Sample Dataset” below).

3. Add a DataSplit node

   Drag Data Split from the Components panel onto the canvas.

   Connect the DataLoader’s right handle to the DataSplit’s left handle.

   Configure DataSplit:

   • Test Split: 20% (default)
   • Random State: 42
   • Stratify: Enable, set column to Survived (or your target)

4. Add a Trainer node

   Drag Trainer onto the canvas and connect from DataSplit.

   Configure Trainer:

   • Mode: Train (default)
   • Model Type: Random Forest Classifier
   • Target Column: Survived

5. Add an Evaluator node

   Drag Evaluator onto the canvas and connect from Trainer.

   No configuration needed—it auto-detects the model type.

6. Run the pipeline

   Click the Run button in the toolbar.

   Watch the Logs tab as each node executes:

   [DataLoader] Loaded titanic.csv: 891 rows, 12 columns
   [DataSplit] Split: 712 train, 179 test (stratified by Survived)
   [Trainer] Training RandomForestClassifier...
   [Trainer] Training complete
   [Evaluator] Accuracy: 0.821, F1: 0.756

7. View results

   Click the Metrics tab to see:

   • Bar chart with Accuracy, Precision, Recall, F1
   • Confusion matrix heatmap

   Click Explain to generate:

   • Feature importance chart
   • SHAP beeswarm plot
   • Partial dependence plots

Sample Dataset

If you don’t have a CSV file, create the Titanic dataset:

import pandas as pd
from sklearn.datasets import fetch_openml

titanic = fetch_openml('titanic', version=1, as_frame=True)
df = titanic.frame
df.to_csv("titanic.csv", index=False)
print(f"Saved titanic.csv with {len(df)} rows")

Or use the Iris dataset for a simpler example:

from sklearn.datasets import load_iris
import pandas as pd

iris = load_iris(as_frame=True)
df = iris.frame
df.to_csv("iris.csv", index=False)
print("Saved iris.csv")

For Iris, set Target Column to target in the Trainer.

Understanding the Results

Metrics Tab

Metric	What It Means
Accuracy	Overall correctness (correct / total)
Precision	When we predict positive, how often are we right?
Recall	Of all actual positives, how many did we find?
F1 Score	Balance of precision and recall

Confusion Matrix

              Predicted
            Died  Survived
Actual Died   98      12
     Survived 21      48

• Diagonal = correct predictions
• Off-diagonal = errors

Explain Section

Click Explain to see why your model makes predictions:

• Feature Importance — Which features matter most
• SHAP Beeswarm — How each feature pushes predictions up/down
• Partial Dependence — How changing a feature affects predictions

Output Panel Tabs

Explore the other tabs at the bottom:

Tab	Purpose
Logs	Execution output and errors
Data Profile	Dataset statistics
Metrics	Model performance charts
Runs	History of all pipeline runs
Models	Registered models with versioning
Trials	Hyperparameter tuning results
Serving	HTTP model server

Save Your Pipeline

Click Save in the toolbar:

1. Enter a name (e.g., “titanic-classifier”)
2. Click Save

Your pipeline is stored locally and appears in the Load dropdown.

Next Steps

Tune Hyperparameters

Use Optuna to find optimal model settings

Explain Predictions

Understand models with SHAP and feature importance

---

Troubleshooting:

• “No Python found” — See Python Setup
• “Module not found: sklearn” — Run pip install scikit-learn
• “Target column not found” — Check column name matches exactly (case-sensitive)
• Pipeline stuck — Check the Logs tab for error messages