def Main():
print("Starting")
global TIMER
TIMER = Timer()
oparser = argparse.ArgumentParser(
description="intelligent crawling with q-learning")
oparser.add_argument(
"--config-file",
dest="configFile",
required=True,
help="Path to config file (containing MySQL login etc.")
oparser.add_argument(
"--save-dir",
dest="saveDir",
default=".",
help=
"Directory that model WIP are saved to. If existing model exists then load it"
)
oparser.add_argument(
"--delete-duplicate-transitions",
dest="deleteDuplicateTransitions",
default=False,
help="If True then only unique transition are used in each batch")
oparser.add_argument(
"--language-pair",
dest="langPair",
required=True,
help="The 2 language we're interested in, separated by ,")
options = oparser.parse_args()
np.random.seed()
np.set_printoptions(formatter={'float': lambda x: "{0:0.1f}".format(x)},
linewidth=666)
sqlconn = MySQL(options.configFile)
hostName = "http://vade-retro.fr/"
#hostName = "http://www.buchmann.ch/"
#hostName = "http://www.visitbritain.com/"
#pickleName = hostName + ".pickle"
env = Env(sqlconn, hostName)
# if os.path.exists(pickleName):
# with open(pickleName, 'rb') as f:
# print("unpickling")
# env = pickle.load(f)
# else:
# env = Env(sqlconn, hostName)
# with open(pickleName, 'wb') as f:
# print("pickling")
# pickle.dump(env,f)
languages = Languages(sqlconn.mycursor)
params = LearningParams(languages, options.saveDir,
options.deleteDuplicateTransitions,
options.langPair)
tf.reset_default_graph()
qns = Qnets(params, env)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as sess:
#with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess:
sess.run(init)
qns.q[0].PrintAllQ(params, env, sess)
#WalkAll(env, params, sess, qns.q[0])
print()
TIMER.Start("Train")
totRewards, totDiscountedRewards = Train(params, sess, saver, env, qns)
TIMER.Pause("Train")
#qn.PrintAllQ(params, env, sess)
#env.WalkAll(params, sess, qn)
Walk(env, sys.maxsize, params, sess, qns.q[0], True)
del TIMER
plt.plot(totRewards)
plt.plot(totDiscountedRewards)
plt.show()
plt.plot(qns.q[0].corpus.losses)
plt.plot(qns.q[1].corpus.losses)
plt.show()
plt.plot(qns.q[0].corpus.sumWeights)
plt.plot(qns.q[1].corpus.sumWeights)
plt.show()
print("Finished")
