About GB_Regressor:

This package provides regressing using gradient boosting technique. There are two options:

1. XGBoost: This is built using XGB libraries
2. PA-GB: This is built in part by using Scikit-learn's DecisionTreeRegressor

Once installed, the user needs to place the raw data into the ./data/raw folder. Through the gateway gb_regressor.py, the application will provide three functionalities:

i. splitdata: The application automatically processes the data and create X_train, X_test, y_train, y_test data tables and saves them in ./data/processed folder.

ii. train: The application provide user options to select betweeen xgb or pa-gb regressors to train. The application will then save the serialized trained model under ./models/ for a provided model name.

iii. predict: The application provide user options to select predictor methods xgb or pa-gb. NOTE: 'xgb' predictor will only with 'xgb' trained models and 'pa-gb' predictor only works with 'pa-gb' trained model. The application utilizes the splited X_test, and y_test data to perform Predictions and evaluate RMSE.

Installation:

In the project's root directory (where this file is), run the followings:

1. Install requirements:

foo@bar pa-gbooster:$ pip install -r requirements.txt

2. To setup, run:

foo@bar pa-gbooster:$ pip install .

How to run:

Note: The feature column headers and target header needs to be defined under Data_Properties class located in:

/gb_regressor/pa_app_config/appconfig.py

1. Place the raw .csv files into:

/data/raw

The program will automatically stitches the files and does the processing.

2. Navigate to the package folder gb_regressor:

foo@bar pa-gbooster:$ cd gb_regressor

3. Run the following command to see the program's option:

foo@bar pa-gbooster/gb_regressor:$ python gb_regressor.py -h

Here are the options:

GB_Regressor - Parsian Asgari's Gradient Booster Regressor

Usage:
    gb_regressor.py splitdata <test-size>
    gb_regressor.py train <trainer-type> <save-model-name> 
    gb_regressor.py predict <predictor-type> <load-model-name>
    gb_regressor.py (-h | --help)

Arguments:
    <test-size>        Choose the fraction size of the test data. Between: 0,10 [default: 0.2]
    <trainer-type>     Choose Gradient Booster Trainer: pa-gb, xgb. [default: xgb]
    <predictor-type>   Choose Gradient Booster Predictor: pa-gb, xgb. [default: xgb]
    <save-model-name>  Serialized model file name to save.
    <load-model-file>  Serialized model file.

Options:
    -h --help          Show this screen.

4. To run, training and testing data sets needs to be made atleast once. If you are running the training and prediction for the first time, please follow the below instructions:

4-1. Splitting Data:

• example: Splitting the data located in the ./data/raw:

foo@bar pa-gbooster/gb_regressor:$ python gb_regressor.py splitdat 0.2

4-2. Training: select either Parsian's Gradient Booster Regressor pa-ga or XGBoost Regressor xgb:

• example: Train the xgb model, serialize and save it as xgb-trained-model

foo@bar pa-gbooster/gb_regressor:$ python gb_regressor.py train xgb xgb-trained-model

4-3. Predicting: select either Parsian's Gradient Booster Predictor pa-ga or XGBoost Predictor xgb:

• example: Using the step above's pre-trained serialized model, lets do some predictions on the X_test data:

foo@bar pa-gbooster/gb_regressor:$ python gb_regressor.py predict xgb xgb-trained-model