This repository contains code for scraping, analyzing, and modeling the odds of the online betting game Aviator. The project includes a full pipeline for collecting historical data, exploring statistical patterns, and training a sequence model (LSTM) to predict future odds based on past behavior.
- Collect and clean historical Aviator odds data
- Perform statistical analysis to uncover structure and potential correlations
- Train a Long Short-Term Memory (LSTM) model to predict upcoming odds
- Investigate whether Aviator odds follow a stochastic pattern or are predictable to some extent
- ~15,000 rounds collected via web scraping
- Each record includes:
timestamp,odds,round_id - Cleaned and stored in CSV format for reproducibility
- LSTM sequence model implemented in PyTorch
- Trained to predict the next odds value given the previous n odds
- Configurable sequence length, learning rate, batch size, and architecture
- Time-series plots of odds distribution and variance
- Rolling statistics and autocorrelation analysis
- Training & validation loss curves
- Predicted vs actual odds visualization
pip install -r requirements.txt
python scraping/scrape.py --output data/aviator_raw.csv
python stats/analyze.py --input data/aviator_clean.csv
python models/train_lstm.py --config configs/train.yaml
python models/evaluate.py --checkpoint checkpoints/model_best.pt