Pre-molded models for classification
Unless you want to run Xgboost on GPU, just follow the guide.
Assuming you have installed python 3.6-7 installed, activate the virtual environment and run pip3 install -r requirements.txt to install the dependencies.
OBS: Note there is no feature engineering in the code. You have to implement your own.
The stacked model uses Lasso as meta model and ElasticNet, KernelRidge and GradientBoosting as base models.
The final predictoin is made averaging the stacked model with a xgboost and lightgbm models. The proportion follows:
Stacked: 0.65
Xgboost: 0.2
Lighgbm: 0.15
To train run python3 stacked_regressor.py passing -p argument indicating the path to the CSV file. Ex:
python3 stacked_regressor.py -p path.csv
The logs will be saved in logs/stacked_regressor folder.
The xgboost models is made to make categorical predictions. There is no needed to take care with the label encoding; the code does itself.
Pay attetion there is no feature engineering. The only thing made is the label encoding. Any column that has categorical data will be changed to numerical data.
Also, note that stratified cross validation will be performed by the algorithm and 10 models will be save for later usage.
The code itself is made to classify two or more labels. Feel free to change the params in config/__init__.py file.
To train run python3 xgb_classifier_cv.py passing -p argument indicating the path to the CSV file. Ex:
python3 xgb_classifier_cv.py -p path.csv
The logs will be save in logs/xgb_classifier folder.
- Stacked classifier
- Downsampling classification