This project is the first version of Track Particles problem and it is part of SPRACE sponsored by Serrapilheira.
To run:
- Clone our repository
- Configure your conda envirotment
- Run
python main.py --config config.json
. It will create a first model with LSTM architecture. If you don't have acces to a gpu, you could use thetest_1_lstm_paralel.ipynb
file to run into Google Colab.
We are using regressions metrics for each hit. Our accuracy is showed for forecas of 5 hit.
[Output] ---Regression Scores---
R_2 statistics (R2) = 0.97
Root Mean Square Error(RMSE) = 0.17
Mean Absolute Error (MAE) = 0.085
RMSE: [0.170]
RMSE features: [0.15, 0.23, 0.09]
R^2 features: [0.97, 0.94, 0.99]
Open the plot_prediction.ipynb file to see the 5 hit predicted.