import tensorflow as tf

import subprocess

import logging

import os

MAX_STEPS = 300000

STEPS_PER_VALIDATION = 1000

STEPS_PER_CHECKPOINT = 20000

TEST_THRESHOLD = 200000

train_params = {

"--steps_per_validation": STEPS_PER_VALIDATION,

"--steps_per_checkpoint": STEPS_PER_CHECKPOINT,

}

if __name__ == "__main__":

logging.basicConfig(level=logging.DEBUG,

format="%(asctime)s %(filename)s[line:%(lineno)d] %(levelname)s %(message)s",

datefmt='%b %d %H:%M')

try:

global_step = tf.contrib.framework.load_variable("model", "global_step")

except:

global_step = 0

logging.info("Training starts with global_step={}. ".format(global_step))

while global_step < MAX_STEPS:

terminate_step = max(global_step + STEPS_PER_CHECKPOINT, TEST_THRESHOLD)

logging.info("Train from {} to {}. ".format(

global_step, terminate_step))

proc = ["python3", "src/summarization.py",

"--max_iter", str(terminate_step)]

for key, val in train_params.items():

proc.append(key)

proc.append(str(val))

subprocess.call(proc)

global_step = terminate_step

subprocess.call(["python3", "script/test.py"])