def train(self, experiences, batch_steps=1, **kwargs):
#state_actions = ssbm.readStateActions(filename)
#feed_dict = feedStateActions(state_actions)
#experiences = util.async_map(ssbm.readStateActions_pickle, filenames)()
experiences = util.deepZip(*experiences)
experiences = util.deepMap(np.array, experiences)
input_dict = dict(
util.deepValues(util.deepZip(self.experience, experiences)))
"""
saved_data = self.sess.run(self.saved_data, input_dict)
handles = [t.handle for t in saved_data]
saved_dict = dict(zip(self.placeholders, handles))
"""
if self.debug:
self.debugGrads(input_dict)
for _ in range(batch_steps):
results = self.sess.run(self.run_dict, input_dict)
summary_str = results['summary']
global_step = results['global_step']
self.writer.add_summary(summary_str, global_step)
def act(self, history, verbose=False):
feed_dict = dict(
util.deepValues(
util.deepZip(
self.input,
ct.vectorizeCTypes(ssbm.SimpleStateAction, history))))
return self.model.act(self.sess.run(self.policy, feed_dict), verbose)