TyreIQ

KO | EN

gitlite — search

#gemini #bash #macos #http #machine-learning #containers #rag #reactjs #chatbot #electron

(★ 9)

No description available.

File Explorer

ZIP Download

- 1_Degradation_Analysis.py
- 2_Pit_Stop_Advisor.py
- 3_Race_Simulator.py
- 4_Driver_Explorer.py
.gitignore
app.py
data_pipeline.py
model_train.py
README.md
requirements.txt
utils.py

# Use via CDN

jsDelivr

jsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.

Version

File

Link

Example

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

> Prototype built for the **TrackShift Innovation Challenge**.  
> Predicts tyre degradation and recommends optimal pit windows using machine learning.

---

## What it does

| Feature | Description |
|---|---|
| **Degradation Curves** | Predicts lap time evolution for Soft / Medium / Hard compounds |
| **Pit Stop Advisor** | AI-generated pit window (Lap X–Y) with time-loss projections |
| **Race Simulator** | Compare stay-out vs pit strategies from any race position |
| **Driver Explorer** | Browse actual lap data, tyre usage, and stint breakdowns |

---

## Quick Start

### 1 — Install dependencies

```bash
pip install -r requirements.txt
```

> **XGBoost** is preferred but the app falls back to scikit-learn's GradientBoostingRegressor automatically if it isn't available.

### 2 — Generate data & train the model

```bash
python data_pipeline.py
python model_train.py
```

This creates:
- `data/race_data.csv` — synthetic 2023 Bahrain GP dataset (8 drivers × 57 laps)
- `models/degradation_model.pkl` — trained regression model
- `models/model_meta.json` — model metrics and pit window metadata

> **Note:** If you have FastF1 installed and an internet connection, `data_pipeline.py` will attempt to pull real 2023 Bahrain GP data first.

### 3 — Launch the dashboard

```bash
streamlit run app.py
```

The app auto-trains the model on first launch if `models/degradation_model.pkl` doesn't exist, so you can skip step 2 if preferred.

---

## Project Structure

```
TyreIQ/
├── app.py              # Streamlit dashboard (4 tabs, F1 dark theme)
├── data_pipeline.py    # Data generation, feature engineering
├── model_train.py      # Model training, prediction helpers
├── requirements.txt
├── data/
│   └── race_data.csv   # Generated on first run
└── models/
    ├── degradation_model.pkl
    └── model_meta.json
```

---

## How the ML works

### Data
Synthetic 2023 Bahrain GP data (8 drivers, 3 compounds, 57 laps per driver) generated with physically-motivated degradation curves including:
- Compound-specific linear + quadratic degradation coefficients
- A tyre "cliff" effect at high age (lap 28/38/50 for S/M/H)
- Fuel load effect (~0.08s/10kg)
- Track temperature deviation from optimal (35°C)

### Features
`tyre_age_laps`, `tyre_age_sq`, `compound_encoded`, `track_temp`, `temp_deviation`, `fuel_load`, `fuel_effect`

### Model
**XGBoost** (or GradientBoostingRegressor fallback) — trained to predict `lap_time`.  
Typical performance on this dataset: **MAE < 0.10s, R² > 0.98**.

### Pit Window Logic
The optimal pit window is the tyre age range where predicted lap time exceeds the fresh-tyre baseline by more than **0.3 seconds/lap**.

---

## Dashboard Tabs

1. **Degradation Analysis** — Side-by-side compound curves + performance loss chart
2. **Pit Stop Advisor** — Live recommendation card with pit window and time-loss delta
3. **Race Simulator** — Stay-out vs pit scenario comparison with total time delta
4. **Driver Data Explorer** — Per-driver lap times, tyre usage, raw data table

---

## Tech Stack

- **Python 3.10+**
- **Streamlit** — dashboard
- **Plotly** — interactive charts
- **XGBoost / scikit-learn** — degradation model
- **Pandas / NumPy** — data processing
- **FastF1** *(optional)* — real F1 data

---

*Built for the TrackShift Innovation Challenge.*

TyreIQ

Owner Repos

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

TyreIQ

Owner Repos

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

Download README