tf_regression_experiment.py

# -*- encoding: utf-8 -*-
from __future__ import print_function
import os
from tf_regression import SimpleModel
import datetime
from multiprocessing import Process


def simulate(bd, ed):
    bd = int(bd)
    ed = int(ed)
    s_date = "%d_%d" % (bd, ed)
    print("train model %s" % s_date)
    sm = SimpleModel()
    X_train, Y_train, _ = sm.load_all_data(bd, ed)
    sm.train_model_tensorflow(X_train, Y_train, s_date)
    sm.save_scaler(s_date)
    sm.load_scaler(s_date)
    test_bd = datetime.date(int(ed/10000), int(ed%10000/100), ed%100) - datetime.timedelta(days=40)
    test_ed = datetime.date(int(ed/10000), int(ed%10000/100), ed%100) + datetime.timedelta(days=40)
    test_bd = int(test_bd.strftime("%Y%m%d"))
    test_ed = int(test_ed.strftime("%Y%m%d"))
    print("Evaluation on %d - %d" % (test_bd, test_ed))
    X_test, Y_test, Data = sm.load_all_data(test_bd, test_ed)
    fname = "../experiments/tensorflow/big/%s/%d_%d.txt" % (s_date, test_bd, test_ed)
    if not os.path.exists("../experiments/tensorflow/big/%s/" % s_date):
        os.makedirs("../experiments/tensorflow/big/%s/" % s_date)
    print("Save results to %s" % fname)
    sm.evaluate_model(X_test, Y_test, Data, s_date, fname)


def simulate_all():
    base_month = 201501
    while base_month <= 201701:
        end_date = datetime.date(int(base_month/100), base_month%100, 1)
        begin_date = datetime.date(int(base_month/100)-5, base_month%100, 1)
        print(begin_date)
        p = Process(target=simulate, args=(begin_date.strftime("%Y%m%d"), end_date.strftime("%Y%m%d")))
        p.start()
        p.join()
        base_month += 1
        if base_month%100 > 12:
            base_month += 88


if __name__ == '__main__':
    simulate_all()