Merge branch 'randomization' into add_preprocess_action - bring in improvements, avoid rebase

bin/HFO

@@ -53,7 +53,11 @@ def launch(cmd, name = 'Unknown', necessary = True, suppressOutput = True):
 def main(args):
   """Sets up the teams, launches the server and monitor, starts the trainer.
   """
-  if args.logging and not os.path.exists(args.logDir):
+  if args.logging:
+    args.hfo_logging = True
+    if not os.path.exists(args.logDir):
+      os.makedirs(args.logDir)
+  elif args.hfo_logging and not os.path.exists(args.logDir):
     os.makedirs(args.logDir)
   num_agents   = args.offenseAgents + args.defenseAgents
   binary_dir   = os.path.dirname(os.path.realpath(__file__))
@@ -77,7 +81,7 @@ def main(args):
                   'server::say_msg_size=%i ' \
                   'server::record_messages=%i' \
                   %(server_port, coach_port, olcoach_port,
-                    args.logging, args.logging, args.logging,
+                    args.logging, args.logging, args.hfo_logging,
                     args.logDir, args.logDir, args.logDir,
                     args.sync, args.fullstate, args.fullstate,
                     args.maxFramesPerTrial, args.numTrials, args.numFrames,
@@ -166,6 +170,8 @@ def parseArgs():
                  'incrementally allocated the following ports.')
   p.add_argument('--no-logging', dest='logging', action='store_false',
                  default=True, help='Disable rcssserver logging.')
+  p.add_argument('--hfo-logging', dest='hfo_logging', action='store_true', default=False,
+                 help="Do .hfo logging even if no other logging")
   p.add_argument('--log-dir', dest='logDir', default='log/',
                  help='Directory to store logs. Default: log/')
   p.add_argument('--record', dest='record', action='store_true',

example/base_2v2.sh

+#!/bin/bash
+
+./bin/HFO --offense-npcs=2 --defense-npcs=2 --trials 20 --headless &
+
+# The magic line
+#   $$ holds the PID for this script
+#   Negation means kill by process group id instead of PID
+trap "kill -TERM -$$" SIGINT
+wait

example/base_long_2v2.sh

+#!/bin/bash
+
+# Be sure to change/remove the seed for different experiments!
+
+./bin/HFO --offense-npcs=2 --defense-npcs=2 --trials 5000 --headless  --seed=1500348586 --no-logging --hfo-logging &
+
+# The magic line
+#   $$ holds the PID for this script
+#   Negation means kill by process group id instead of PID
+trap "kill -TERM -$$" SIGINT
+wait

example/defense_2v3.sh

+#!/bin/bash
+
+# HAS TO BE RUN FROM EXAMPLE DIR DUE TO hand_coded_defense_agent CONFIG!
+
+../bin/HFO --offense-npcs=2 --defense-agents=1 --defense-npcs=2 --trials 20 --headless --port=7000 &
+# The below sleep period is needed to avoid the agent connecting in before the
+# Trainer.py script gets the base/Helios goalie connected in; if that happens,
+# the agent gets assigned unum 1 and there is a mixup in which agent is
+# supposed to be the goalie (some portions of the various programs go by unum,
+# others go by a goalie flag).
+sleep 15
+./hand_coded_defense_agent &> agent1.txt &
+sleep 5
+
+# The magic line
+#   $$ holds the PID for this script
+#   Negation means kill by process group id instead of PID
+trap "kill -TERM -$$" SIGINT
+wait
\ No newline at end of file

example/defense_long_2v2.sh

@@ -2,7 +2,9 @@
 
 # HAS TO BE RUN FROM EXAMPLE DIR DUE TO hand_coded_defense_agent CONFIG!
 
-../bin/HFO --offense-npcs=2 --defense-agents=1 --defense-npcs=1 --trials 5000 --headless --port=7000 --seed 1500310928 --no-logging &
+# Change to a new seed for different experiments!
+
+../bin/HFO --offense-npcs=2 --defense-agents=1 --defense-npcs=1 --trials 5000 --headless --port=7000 --seed 1500348586 --no-logging &
 sleep 15
 ./hand_coded_defense_agent &> agent1.txt &
 sleep 5

example/defense_python_long_2v2.sh

 #!/bin/bash
 
-./bin/HFO --offense-npcs=2 --defense-agents=1 --defense-npcs=1 --trials 5000 --headless --seed 1500310928 --no-logging &
+# Change to a different seed for different experiments!
+
+./bin/HFO --offense-npcs=2 --defense-agents=1 --defense-npcs=1 --trials 5000 --headless --seed 1500348586 --no-logging &
 sleep 15
 ./example/hand_coded_defense_agent.py &> agent1.txt &
 sleep 5

example/hand_coded_defense_agent.cpp

@@ -66,7 +66,7 @@ bool is_in_open_area(double pos_x, double pos_y) {
         }
 }
 
-action_with_params get_defense_action(const std::vector<float>& state_vec, double no_of_opponents, double numTMates) {
+action_with_params get_defense_action(const std::vector<float>& state_vec, double no_of_opponents, int numTMates) {
         int size_of_vec = 10 + 6*numTMates + 3*no_of_opponents;
         if (size_of_vec != state_vec.size()) {
                 std :: cout <<"Invalid Feature Vector / Check the number of teammates/opponents provided";
@@ -221,7 +221,7 @@ int main(int argc, char** argv) {
     // Get the vector of state features for the current state
       const vector<float>& feature_vec = hfo.getState();
       if (random == 0) {
-              action_with_params a = get_defense_action(feature_vec, 2,1);
+              action_with_params a = get_defense_action(feature_vec, 2, hfo.getNumTeammates());
          // std::cout << a.action << a.param;
          if (a.action == hfo :: MARK_PLAYER || a.action == hfo::TACKLE) {
                   hfo.act(a.action, a.param);

example/hand_coded_defense_agent.py

@@ -7,8 +7,6 @@ keeper/goalie.
 from __future__ import print_function
 # encoding: utf-8
 
-#MODIFIED#
-
 # First Start the server: $> bin/start.py
 import argparse
 import itertools
@@ -83,9 +81,17 @@ def get_sorted_opponents(state_vec, num_opponents, num_teammates, pos_x, pos_y):
 def is_in_open_area(pos_x, ignored_pos_y):
   return pos_x >= params['OPEN_AREA_HIGH_LIMIT_X']
 
-def do_defense_action(state_vec, hfo_env, episode,
+def add_num_times(action, main_dict, opt_dict=None):
+  main_dict[action] += 1
+  if opt_dict:
+    opt_dict[action] += 1
+  return action
+
+def do_defense_action(state_vec, hfo_env,
                       num_opponents, num_teammates,
-                      old_ball_pos_x, old_ball_pos_y):
+                      old_ball_pos_x, old_ball_pos_y,
+                      num_times_overall, num_times_kickable,
+                      misc_tracked):
   """Figures out and does the (hopefully) best defense action."""
   min_vec_size = 10 + (6*num_teammates) + (3*num_opponents)
   if (len(state_vec) < min_vec_size):
@@ -110,19 +116,19 @@ def do_defense_action(state_vec, hfo_env, episode,
   ball_sorted_list = get_sorted_opponents(state_vec, num_opponents, num_teammates,
                                           pos_x=ball_pos_x, pos_y=ball_pos_y)
   if not ball_sorted_list: # unknown opponent positions/unums
-    print("No known opponent locations " +
-          "(episode {0:d}; btg {1!r}; ".format(episode,ball_toward_goal) +
+    print("No known opponent locations (btg {0!r}; bng {1!r}; ".format(ball_toward_goal,
+                                                                       ball_nearer_goal) +
           "ball xy {0:n}, {1:n}; ball old xy {2:n}, {3:n})".format(ball_pos_x,
                                                                    ball_pos_y,
                                                                    old_ball_pos_x,
                                                                    old_ball_pos_y))
-    if ball_toward_goal and (not is_in_open_area(ball_pos_x, ball_pos_y)):
+    if ball_toward_goal:
       if ball_nearer_goal:
-        hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+        hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
       else:
-        hfo_env.act(hfo.INTERCEPT)
+        hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall))
     else:
-      hfo_env.act(hfo.MOVE)
+      hfo_env.act(add_num_times(hfo.MOVE,num_times_overall))
     return
 
   goal_sorted_list = get_sorted_opponents(state_vec, num_opponents, num_teammates,
@@ -139,96 +145,96 @@ def do_defense_action(state_vec, hfo_env, episode,
                                     ball_sorted_list[0][1],
                                     ball_sorted_list[0][2], ball_sorted_list[0][3])
 
+  agent_to_ball_dist = get_dist_normalized(agent_pos_x, agent_pos_y,
+                                           ball_pos_x, ball_pos_y)
+
   if state_vec[5] > 0: # kickable distance of player
+    misc_tracked['max_kickable_dist'] = max(agent_to_ball_dist,misc_tracked['max_kickable_dist'])
     if is_tackleable_opp:
-      hfo_env.act(hfo.MOVE) # will do tackle
+      hfo_env.act(add_num_times(hfo.MOVE,num_times_overall,num_times_kickable)) # will do tackle
     elif ball_nearer_goal:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall,num_times_kickable))
     elif ball_toward_goal:
-      hfo_env.act(hfo.INTERCEPT)
+      hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall,num_times_kickable))
     else:
-      hfo_env.act(hfo.GO_TO_BALL)
+      hfo_env.act(add_num_times(hfo.GO_TO_BALL,num_times_overall,num_times_kickable))
     return
 
-  agent_to_ball_dist = get_dist_normalized(agent_pos_x, agent_pos_y,
-                                           ball_pos_x, ball_pos_y)
-
   if goal_sorted_list[0][0] != ball_sorted_list[0][0]:
     if is_in_open_area(ball_sorted_list[0][2],
                        ball_sorted_list[0][3]) and is_in_open_area(goal_sorted_list[0][2],
                                                                    goal_sorted_list[0][3]):
       if ball_sorted_list[0][1] < params['LOW_KICK_DIST']:
-        hfo_env.act(hfo.MARK_PLAYER, goal_sorted_list[0][0])
+        hfo_env.act(add_num_times(hfo.MARK_PLAYER,num_times_overall),
+                    goal_sorted_list[0][0])
       elif agent_to_ball_dist < ball_sorted_list[0][1]:
         if ball_nearer_goal:
-          hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+          hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
         elif ball_toward_goal:
-          hfo_env.act(hfo.INTERCEPT)
+          hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall))
         else:
-          hfo_env.act(hfo.GO_TO_BALL)
+          hfo_env.act(add_num_times(hfo.GO_TO_BALL,num_times_overall))
       else:
-        hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+        hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
     elif ball_sorted_list[0][1] >= params['KICK_DIST']:
       if agent_to_ball_dist < ball_sorted_list[0][1]:
         if ball_nearer_goal:
-          hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+          hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
         elif ball_toward_goal:
-          hfo_env.act(hfo.INTERCEPT)
+          hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall))
         else:
-          hfo_env.act(hfo.GO_TO_BALL)
+          hfo_env.act(add_num_times(hfo.GO_TO_BALL,num_times_overall))
       else:
-        hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+        hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
     elif is_tackleable_opp and (not is_in_open_area(ball_sorted_list[0][2],
                                                     ball_sorted_list[0][3])):
-      hfo_env.act(hfo.MOVE)
-##    elif is_in_open_area(ball_sorted_list[0][2],ball_sorted_list[0][3]):
-##      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL) # why not MARK_PLAYER for the one that is not in the open area?
+      hfo_env.act(add_num_times(hfo.MOVE,num_times_overall))
     elif ball_sorted_list[0][1] < (1*params['LOW_KICK_DIST']):
-      hfo_env.act(hfo.MARK_PLAYER, goal_sorted_list[0][0])
+      hfo_env.act(add_num_times(hfo.MARK_PLAYER,num_times_overall),
+                  goal_sorted_list[0][0])
     else:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
     return
 
   if is_in_open_area(ball_sorted_list[0][2],ball_sorted_list[0][3]):
     if ball_sorted_list[0][1] < params['KICK_DIST']:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
     elif agent_to_ball_dist < params['KICK_DIST']:
       if ball_nearer_goal:
-        hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+        hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
       elif ball_toward_goal:
-        hfo_env.act(hfo.INTERCEPT)
+        hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall))
       else:
-        hfo_env.act(hfo.GO_TO_BALL)
+        hfo_env.act(add_num_times(hfo.GO_TO_BALL,num_times_overall))
     elif is_tackleable_opp:
-      hfo_env.act(hfo.MOVE)
+      hfo_env.act(add_num_times(hfo.MOVE,num_times_overall))
     else:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
   else:
     if ball_sorted_list[0][1] >= max(params['KICK_DIST'],agent_to_ball_dist):
       if ball_nearer_goal:
-        hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+        hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
       elif ball_toward_goal:
-        hfo_env.act(hfo.INTERCEPT)
+        hfo_env.act(add_num_times(hfo.INTERCEPT,num_times_overall))
       else:
-        hfo_env.act(hfo.GO_TO_BALL)
+        hfo_env.act(add_num_times(hfo.GO_TO_BALL,num_times_overall))
     elif ball_sorted_list[0][1] >= params['KICK_DIST']:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
     elif is_tackleable_opp:
-      hfo_env.act(hfo.MOVE)
+      hfo_env.act(add_num_times(hfo.MOVE,num_times_overall))
     else:
-      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
+      hfo_env.act(add_num_times(hfo.REDUCE_ANGLE_TO_GOAL,num_times_overall))
   return
 
 def do_random_defense_action(state, hfo_env):
   if state[5] > 0: # kickable
-    if random.random() < 0.5:
-      hfo_env.act(hfo.INTERCEPT)
+    hfo_env.act(hfo.MOVE)
+  else:
+    if random.random() < 0.25:
+      hfo_env.act(hfo.REDUCE_ANGLE_TO_GOAL)
     else:
       hfo_env.act(hfo.MOVE)
-  else:
-    hfo_env.act(random.choose(hfo.MOVE,hfo.DEFEND_GOAL,
-                              hfo.REDUCE_ANGLE_TO_GOAL,hfo.REDUCE_ANGLE_TO_GOAL,
-                              hfo.GO_TO_BALL,hfo.INTERCEPT))
+
   return
 
 def main():
@@ -267,6 +273,12 @@ def main():
   my_unum = hfo_env.getUnum()
   assert ((my_unum > 1) and (my_unum <= 11)), "Bad unum {!r}".format(my_unum)
   print("My unum is {0:d}".format(my_unum))
+  num_times_overall = {}
+  num_times_kickable = {}
+  for action in range(hfo.NUM_HFO_ACTIONS):
+    num_times_overall[action] = 0
+    num_times_kickable[action] = 0
+  misc_tracked = {'max_kickable_dist':0}
   for episode in itertools.count():
     old_ball_pos_x = -1
     old_ball_pos_y = 0
@@ -283,9 +295,12 @@ def main():
       if (args.epsilon > 0) and (random.random() < args.epsilon):
         do_random_defense_action(state, hfo_env)
       else:
-        do_defense_action(state_vec=state, hfo_env=hfo_env, episode=episode,
+        do_defense_action(state_vec=state, hfo_env=hfo_env,
                           num_opponents=numOpponents, num_teammates=numTeammates,
-                          old_ball_pos_x=old_ball_pos_x, old_ball_pos_y=old_ball_pos_y)
+                          old_ball_pos_x=old_ball_pos_x, old_ball_pos_y=old_ball_pos_y,
+                          num_times_overall=num_times_overall,
+                          num_times_kickable=num_times_kickable,
+                          misc_tracked=misc_tracked)
       old_ball_pos_x=state[3]
       old_ball_pos_y=state[4]
       status=hfo_env.step()
@@ -293,6 +308,15 @@ def main():
 
     # Quit if the server goes down
     if status == hfo.SERVER_DOWN:
+      for action in range(hfo.NUM_HFO_ACTIONS):
+        if num_times_overall[action]:
+          print("Overall times {0!s}: {1:d}".format(hfo_env.actionToString(action),
+                                                    num_times_overall[action]))
+      for action in range(hfo.NUM_HFO_ACTIONS):
+        if num_times_kickable[action]:
+          print("Kickable times {0!s}: {1:d}".format(hfo_env.actionToString(action),
+                                                     num_times_kickable[action]))
+      print("Max kickable dist: {0:n}".format(misc_tracked['max_kickable_dist']))
       hfo_env.act(hfo.QUIT)
       exit()