train.py

#SCRIPT FOR TRAINING THE MODELS

import ipdb
import numpy as np
from data import HOSPData,UKDALEData
import matplotlib.pyplot as plt
from modelGen import modelGen
import argparse
import warnings
warnings.filterwarnings("ignore")
from models.utils import reset_seeds
from address import get_model_path
from knockknock import teams_sender



reset_seeds(seed_value=39)


def main(args):
    DATASET        = args.dataset
    TRAINING_HOUSE = args.trainHouse
    TEST_HOUSE     = args.testHouse
    TRAINING_DATES = args.trainDates
    TEST_DATES     = args.testDates
    MODELS         = args.models
    SEQ_LEN        = args.seqLen     
    EPOCHS         = args.epochs   

    # Reading data
    if DATASET == "UKDALE":
        dataGen = UKDALEData(path="./data/")
        trainMain,trainTargets, trainStates = dataGen.get_train_sequences(houses = TRAINING_HOUSE, start = TRAINING_DATES[0],end=TRAINING_DATES[1])
        testMain,testTargets, testStates = dataGen.get_train_sequences(houses = TEST_HOUSE, start=TEST_DATES[0],end=TEST_DATES[1],norm = False)
        app_data = dataGen.get_app_data()
    else:
        dataGen = HOSPData(path="./data/")
        trainMain,trainTargets, trainStates = dataGen.get_train_sequences(start = TRAINING_DATES[0],end=TRAINING_DATES[1])
        testMain,testTargets, testStates = dataGen.get_test_sequences( start=TEST_DATES[0],end=TEST_DATES[1],norm=False)
        app_data = dataGen.get_app_data()
    # Data dict
    data= {"X_train":trainMain,
            "Y_train":trainTargets,
            "Y_states":trainStates,
            "Z_train":trainStates,
            "X_test":testMain,
            "Y_test":testTargets,
            "Z_test":testStates,
            "app_data":dataGen.get_app_data(),
            "main_data": dataGen.get_main_mean_std()
            }  
    
    print(f"EL modelo es {MODELS}")
    for modelID in MODELS:
        params = {"epochs":EPOCHS,"sequence_length":SEQ_LEN}
        model = modelGen(modelID,data,params)
        model.train()
        model.store(get_model_path(DATASET,modelID))
        del model


if __name__ == "__main__":
    # Defining program params
    parser = argparse.ArgumentParser(description="Script to train the UKDALE models.",
                                    formatter_class=argparse.ArgumentDefaultsHelpFormatter)

    parser.add_argument("-dataset",    action="store", help="selected dataset: UKDALE or HOSP ",default="UKDALE")
    parser.add_argument("-trainHouse", action="store", help="train house for UKDALE", type = int,  default=1)
    parser.add_argument("-testHouse",  action="store", help="test house for UKDALE" , type = int,  default=1)
    parser.add_argument("-trainDates", action="store", help="Dates for training",     nargs='+', default=["2015-01-01","2016-01-01"])
    parser.add_argument("-testDates",  action="store", help="Dates for test",        nargs='+', default=["2016-01-01","2016-07-01"])
    parser.add_argument("-models",     action="store", help="Models to be trained",   nargs='+', default=["FCNdAE","multiFCNdAE","multiUNET"])
    parser.add_argument("-seqLen",    action="store", help="Sequence length ",       type = int, default= 500)
    parser.add_argument("-epochs",    action="store", help="Training epochs ",       type = int, default= 100)

    args = parser.parse_args()
    main(args)