Haxxsh/gdgc-datathon-data

---
license: mit
task_categories:
- tabular-regression
tags:
- motorsport
- formula-racing
- lap-time
- tabular
- gdgc-datathon
language:
- en
size_categories:
- 100K<n<1M
---

# GDGC Datathon 2025 - Formula Racing Lap Time Dataset

Dataset for predicting Formula racing lap times, used in the GDGC Datathon 2025 competition.

## Dataset Description

This dataset contains historical Formula racing data with various features related to circuits, weather conditions, rider/driver performance, and race configurations. The goal is to predict `Lap_Time_Seconds`.

### Dataset Summary

| Split | Samples | Size |
|-------|---------|------|
| Train | 734,002 | 124 MB |
| Test | 195,001 | 51 MB |

## Dataset Structure

### Files

```
data/
├── train.csv # Training data with target variable
└── test.csv # Test data for predictions
```

### Features

| Column | Description | Type |
|--------|-------------|------|
| `id` | Unique identifier | int |
| `Unique ID` | Alternative unique ID | int |
| `Rider_ID` | Rider/driver identifier | int |
| `Formula_category_x` | Racing formula category | categorical |
| `Len_Circuit_inkm` | Circuit length in kilometers | float |
| `Laps` | Number of laps in the race | int |
| `Start_Position` | Starting grid position | int |
| `Formula_Avg_Speed_kmh` | Average speed in km/h | float |
| `Formula_Track_Condition` | Track condition rating | categorical |
| `Humidity_%` | Humidity percentage | float |
| `Tire_Compound` | Type of tire compound used | categorical |
| `Penalty` | Penalty time/status | float |
| `Champ_Points` | Championship points | float |
| `Champ_Position` | Championship standing position | int |
| `Session` | Race session type | categorical |
| `race_year` | Year of the race | int |
| `seq` | Sequence number | int |
| `position` | Final position | int |
| `points` | Points earned | float |
| `Formula_shortname` | Short name of formula | categorical |
| `circuit_name` | Name of the circuit | categorical |
| `Corners_in_Lap` | Number of corners per lap | int |
| `Tire_Degradation_Factor_per_Lap` | Tire degradation rate | float |
| `Pit_Stop_Duration_Seconds` | Pit stop time in seconds | float |
| `Ambient_Temperature_Celsius` | Air temperature | float |
| `Track_Temperature_Celsius` | Track surface temperature | float |
| `weather` | Weather condition | categorical |
| `track` | Track identifier | categorical |
| `air` | Air condition metric | float |
| `ground` | Ground condition metric | float |
| `starts` | Number of race starts | int |
| `finishes` | Number of race finishes | int |
| `with_points` | Races finished with points | int |
| `podiums` | Number of podium finishes | int |
| `wins` | Number of wins | int |
| `Lap_Time_Seconds` | **Target variable** - Lap time in seconds | float |

### Target Variable Statistics

| Metric | Value |
|--------|-------|
| Count | 734,002 |
| Mean | 89.997 s |
| Std | 11.532 s |
| Min | 70.001 s |
| 25% | 79.989 s |
| 50% (Median) | 89.970 s |
| 75% | 99.914 s |
| Max | 109.999 s |

The target distribution is **nearly symmetric** with mean ≈ median, indicating no significant skew.

## Usage

### Loading with Pandas

```python
import pandas as pd

# Load training data
train_df = pd.read_csv("train.csv")
print(f"Training samples: {len(train_df)}")

# Load test data
test_df = pd.read_csv("test.csv")
print(f"Test samples: {len(test_df)}")

# Separate features and target
X = train_df.drop(columns=["Lap_Time_Seconds", "id"])
y = train_df["Lap_Time_Seconds"]
```

### Loading from Hugging Face

```python
from huggingface_hub import hf_hub_download
import pandas as pd

# Download files
train_path = hf_hub_download(
repo_id="Haxxsh/gdgc-datathon-data",
filename="train.csv",
repo_type="dataset"
)

test_path = hf_hub_download(
repo_id="Haxxsh/gdgc-datathon-data",
filename="test.csv",
repo_type="dataset"
)

# Load into pandas
train_df = pd.read_csv(train_path)
test_df = pd.read_csv(test_path)
```

### With Datasets Library

```python
from datasets import load_dataset

dataset = load_dataset("Haxxsh/gdgc-datathon-data")
```

## Trained Models

Pre-trained models for this dataset are available at:
- **Models:** [Haxxsh/gdgc-datathon-models](https://huggingface.co/Haxxsh/gdgc-datathon-models)
- **Training Code:** [ezylopx5/DATATHON](https://github.com/ezylopx5/DATATHON)

## Evaluation Metric

The primary evaluation metric is **RMSE** (Root Mean Squared Error):

```python
from sklearn.metrics import mean_squared_error
import numpy as np

rmse = np.sqrt(mean_squared_error(y_true, y_pred))
```

## Data Preprocessing Tips

1. **Handle categorical features:** Use label encoding or one-hot encoding for columns like `weather`, `circuit_name`, `Tire_Compound`
2. **Feature scaling:** Normalize numerical features for certain models
3. **Missing values:** Check for and handle any missing values appropriately
4. **Feature engineering:** Consider creating interaction features or aggregations

## License

MIT License

## Citation

```bibtex
@dataset{gdgc-datathon-2025-data,
author = {Haxxsh},
title = {GDGC Datathon 2025 - Formula Racing Lap Time Dataset},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/Haxxsh/gdgc-datathon-data}
}
```

## Acknowledgments

- GDGC Datathon 2025 organizers
- Formula racing data providers