Train a Classifier

This tutorial walks through building a complete classification pipeline using the Iris dataset. You’ll learn how to:

• Load and preview data
• Configure a classifier
• Evaluate model performance
• Interpret results

Time to complete: ~10 minutes

Prerequisites

• MLOps Desktop installed
• Python packages: pip install scikit-learn pandas

Prepare the Dataset

First, create a sample dataset. Open Terminal and run:

from sklearn.datasets import load_iris
import pandas as pd

iris = load_iris(as_frame=True)
df = iris.frame
df.to_csv("~/Desktop/iris.csv", index=False)
print(f"Saved iris.csv with {len(df)} rows")

This creates iris.csv on your Desktop with 150 samples and 5 columns:

• sepal length (cm), sepal width (cm), petal length (cm), petal width (cm) — features
• target — class label (0=setosa, 1=versicolor, 2=virginica)

Build the Pipeline

1. Create a new pipeline

   Open MLOps Desktop. You’ll see an empty canvas.

2. Add a DataLoader node

   Click + Add Node → DataLoader.

   Click the node to select it, then in the properties panel:

   • Click Browse and select iris.csv
   • The preview shows your data columns

3. Add a Trainer node

   Click + Add Node → Trainer.

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

4. Configure the Trainer

   Click the Trainer node. In the properties panel:

   Setting	Value
   Model Type	Random Forest Classifier
   Target Column	target
   Test Size	0.2
   Random State	42

   Random Forest settings:

   Parameter	Value	Why
   n_estimators	100	Number of trees (more = better, slower)
   max_depth	10	Limit tree depth to prevent overfitting
   min_samples_split	2	Minimum samples to split a node

5. Add an Evaluator node

   Click + Add Node → Evaluator.

   Connect the Trainer’s right handle to the Evaluator’s left handle.

6. Run the pipeline

   Click Run in the toolbar.

   Watch the output panel as each node executes:

   [DataLoader] Loaded iris.csv: 150 rows, 5 columns
   [Trainer] Training Random Forest on 120 samples...
   [Trainer] Training complete
   [Evaluator] Accuracy: 0.967 (29/30 correct)

Understanding Classification Metrics

After running, the Evaluator shows several metrics:

Accuracy

The simplest metric: percentage of correct predictions.

Accuracy = Correct Predictions / Total Predictions
         = 29 / 30
         = 0.967 (96.7%)

Precision, Recall, F1

For each class:

Metric	Formula	Interpretation
Precision	TP / (TP + FP)	“Of all predicted positives, how many were correct?”
Recall	TP / (TP + FN)	“Of all actual positives, how many did we find?”
F1 Score	2 × (P × R) / (P + R)	Harmonic mean of precision and recall

Tip

• High precision, low recall → Conservative predictions (fewer false positives)
• Low precision, high recall → Aggressive predictions (fewer false negatives)
• F1 balances both

Confusion Matrix

A 3×3 matrix showing predictions vs actual labels:

              Predicted
            0    1    2
Actual  0 [ 10   0    0 ]  ← Setosa (perfect!)
        1 [  0   9    1 ]  ← Versicolor (1 misclassified as Virginica)
        2 [  0   0   10 ]  ← Virginica (perfect!)

Diagonal values = correct predictions. Off-diagonal = errors.

Try Different Models

Experiment with other classifiers:

Model	Best For
Logistic Regression	Linear decision boundaries, interpretable
Random Forest	Complex relationships, handles outliers
Gradient Boosting	Maximum accuracy, slower training
SVM	High-dimensional data, binary classification

To switch models:

1. Click the Trainer node
2. Change Model Type
3. Click Run again

Compare accuracy across models to find the best one for your data.

Save Your Pipeline

Click Save and name it “iris-classifier”.

Your pipeline is now saved and can be reloaded anytime from the Load dropdown.

Next Steps

Tune Hyperparameters

Find optimal model settings with Optuna

Explain Predictions

Understand why your model makes predictions

---

Troubleshooting:

• “Column ‘target’ not found” — Check your CSV has a target column, or select the correct column name
• Low accuracy — Try increasing n_estimators or changing the model type
• “Not enough samples” — Ensure test_size leaves enough data for training (at least 50+ samples)