training.py

"""
Python script meant to train an ML model.
"""
import pandas as pd
import os
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
import logging
import json
import sys
from joblib import dump
from common_functions import preprocess_data

logging.basicConfig(stream=sys.stdout, level=logging.INFO)


with open('config.json', 'r') as f:
    """
    Load config.json and get path variables.
    """
    config = json.load(f) 

dataset_csv_path = os.path.join(config['output_folder_path']) 
model_path = os.path.join(config['output_model_path']) 


def train_model():
    """
    Function for training the model.
    """
    df = pd.read_csv(os.path.join(dataset_csv_path, "finaldata.csv"))
    df_x, df_y, encoder = preprocess_data(df, None)
    
    x_train, x_test, y_train, y_test = train_test_split(df_x, df_y, test_size=0.20)

    # Use this logistic regression for training
    model = LogisticRegression(
        C=1.0,
        class_weight=None,
        dual=False,
        fit_intercept=True,
        intercept_scaling=1,
        l1_ratio=None,
        max_iter=100,
        multi_class='ovr',
        n_jobs=None,
        penalty='l2',
        random_state=0,
        solver='liblinear',
        tol=0.0001,
        verbose=0,
        warm_start=False
    )
    
    # Fit the logistic regression to your data
    model.fit(x_train, y_train)
    
    print(model.score(x_train, y_train))
    print(model.score(x_test, y_test))
    
    # Write the trained model to your workspace in a file called trainedmodel.pkl
    dump(model, os.path.join(model_path, "trainedmodel.pkl"))
    dump(encoder, os.path.join(model_path, "encoder.pkl"))


if __name__ == "__main__":
    logging.info("Running Training!")
    train_model()
    logging.info("Artifacts output written in practicemodels/trainedmodel.pkl")
    logging.info("Artifacts output written in practicemodels/encoder.pkl")