- Data : https://www.kaggle.com/competitions/competitive-data-science-predict-future-sales
- Algorithm : Light GBM
- Goal : below RMSE 0.9 Model
Training (command): python train.py
Training (mlflow and docker)
mlflow run -e main .
Docker Image Build: docker build -t mlflow-docker-futuresales -f Dockerfile .
Run Docker Image: mlflow run docker
Light GBM
- Hyperparameters
Hyperparameters | Value |
---|---|
metric | rmse |
num_leaves | 255 |
learning_rate | 0.005 |
feature_fraction | 0.75 |
bagging_fraction | 0.75 |
bagging_freq | 5 |
force_col_wise | True |
- Metric
Data | RMSE |
---|---|
Validation | 0.88399 |
Public Test | 0.87319 |