10.2. Getting Started (GUI)

This guide walks you through training your first model using Edge AI Studio Model Composer’s graphical interface.

10.2.1. Prerequisites

Before starting:

  1. TI account (create at https://www.ti.com)

  2. Modern web browser (Chrome, Firefox, Edge recommended)

  3. Dataset ready (or use built-in example)

10.2.2. Step 1: Access Model Composer

  1. Navigate to https://dev.ti.com/edgeaistudio/

  2. Log in with your TI credentials

  3. From the dashboard, select Model Composer

  4. Click Create New Project

10.2.3. Step 2: Create a Project

Project Setup:

  1. Enter a project name (e.g., “Arc Fault Detection”)

  2. Select task type: Time Series Classification

  3. Select target device: F28P55 (or your target)

  4. Click Create

You’ll be taken to the project dashboard.

10.2.4. Step 3: Upload Dataset

Option A: Use Example Dataset

  1. Click Datasets in the sidebar

  2. Click Add Dataset

  3. Select Use Example Dataset

  4. Choose “DC Arc Fault” from the list

  5. Click Import

Option B: Upload Your Own

  1. Click Datasets → Add Dataset

  2. Select Upload Custom Dataset

  3. Drag and drop your ZIP file or click to browse

  4. Wait for upload and validation

  5. Preview data to verify

Dataset Format:

Your ZIP should contain:

my_dataset.zip
├── annotations.yaml
└── classes/
    ├── class_a/
    │   ├── sample1.csv
    │   └── sample2.csv
    └── class_b/
        └── ...

10.2.5. Step 4: Configure Feature Extraction

  1. Click Feature Extraction in sidebar

  2. Select a preset or customize:

    • Preset: Choose from dropdown (e.g., “FFT 1024 Input, 256 Features”)

    • Custom: Configure individual parameters

  3. Key settings:

    • Input Size: Samples per inference window

    • Transform: FFT, Raw, Wavelet

    • Features: Number of output features

    • Frames: Temporal context

  4. Click Save Configuration

Tip: Start with a preset; customize later if needed.

10.2.6. Step 5: Configure Training

  1. Click Training in sidebar

  2. Configure model settings:

    Model Selection:

    • Select model size (1k, 2k, 4k, etc.)

    • NPU devices show NPU-compatible options

    Training Parameters:

    • Epochs: 20-50 (start with 20)

    • Batch Size: 256 (default)

    • Learning Rate: 0.001 (default)

    Quantization:

    • Enable for NPU devices

    • Select INT8 (recommended)

  3. Click Save Configuration

10.2.7. Step 6: Start Training

  1. Click Train Model button

  2. Confirm your settings in the dialog

  3. Click Start Training

During Training:

  • Progress bar shows completion

  • Loss graph updates in real-time

  • Training log shows detailed status

  • Estimated time remaining displayed

Training Complete:

  • Success message appears

  • Results summary shown

  • Model ready for analysis

10.2.8. Step 7: Analyze Results

After training, review model performance:

Accuracy Tab:

  • Overall accuracy percentage

  • Per-class accuracy breakdown

  • F1 scores

Confusion Matrix Tab:

  • Visual confusion matrix

  • Click cells for sample details

  • Identify which classes are confused

ROC Curves Tab:

  • One-vs-rest ROC for each class

  • AUC scores

  • Threshold selection helper

Sample Viewer Tab:

  • View individual predictions

  • See correct and incorrect samples

  • Understand model behavior

10.2.9. Step 8: Evaluate on Test Set

  1. Click Testing in sidebar

  2. Select test dataset (holdout data)

  3. Click Run Evaluation

  4. Review test set metrics

Compare test metrics to training metrics:

  • Similar = Good generalization

  • Test much lower = Possible overfitting

10.2.10. Step 9: Export Model

  1. Click Export in sidebar

  2. Choose export format:

    • CCS Project: Complete project template

    • Artifacts Only: Just model files

    • Configuration: YAML config for CLI

  3. Click Download

  4. Extract the ZIP file

See Exporting Models for detailed export instructions.

10.2.11. Using Exported Model

After export, you’ll have files for CCS:

export/
├── mod.a                    # Model library
├── mod.h                    # Header file
├── feature_extraction.c    # Feature code
└── example_project/        # CCS project (if selected)

See CCS Integration Guide for deployment steps.

10.2.12. GUI Tips and Tricks

Keyboard Shortcuts:

  • Ctrl+S: Save current configuration

  • Ctrl+Z: Undo last change

  • Ctrl+Enter: Start training

Dataset Preview:

  • Click any sample in dataset view

  • View raw data and labels

  • Check for data quality issues

Training History:

  • All training runs are saved

  • Compare runs in History tab

  • Resume from any previous run

Quick Actions:

  • Duplicate project: Copy with one click

  • Reset configuration: Return to defaults

  • Export configuration: Share settings

10.2.13. Troubleshooting

Dataset Upload Fails:

  • Check file size limits

  • Verify ZIP structure

  • Ensure valid CSV format

Training Doesn’t Start:

  • Check browser console for errors

  • Refresh page and retry

  • Verify dataset is properly loaded

Slow Training:

  • Reduce batch size

  • Use smaller model

  • Check internet connection (cloud)

Export Download Fails:

  • Check browser download settings

  • Try different browser

  • Contact support if persists

10.2.14. What’s Next?

After completing your first model:

  1. Try your own data: Upload custom dataset

  2. Experiment with settings: Different models, features

  3. Compare results: Use training history

  4. Deploy to device: Follow CCS integration guide

10.2.15. Additional Resources