test bot advisor on eval mode - fixed

scripts/eval_bot.py

@@ -3,13 +3,25 @@
 """
 Evaluate the success rate of the bot
 This script is used for testing/debugging purposes
+
+Examples of usage:
+- Run the bot on the GoTo level 10 times (seeds 9 to 18)
+eval_bot.py --level GoTo --num_runs 10 --seed 9
+- for all levels, 100 times, run a Random(seed 0) agent for len(episode)/3 steps before running the bot:
+eval_bot.py --advise_mode --num_runs 100
+- for all levels, 500 times, during the first 10 steps, choose action form a Random(seed 9) agent with proba .9 or
+ optimal (from bot) with proba .1, then continue with optimal bot actions:
+eval_boy.py --advise_mode --bad_action_proba .8 --non_optimal_steps 10 --random_agent_seed 9
+
 """
 
 import random
 import time
 from optparse import OptionParser
 from babyai.levels import level_dict
-from babyai.bot import Bot
+from babyai.bot import Bot, DisappearedBoxError
+from babyai.utils.agent import ModelAgent, RandomAgent
+import numpy as np
 
 level_list = [
     'OpenRedDoor',
@@ -39,6 +51,34 @@
     "--level",
     default=None
 )
+parser.add_option(
+    "--advise_mode",
+    action='store_true',
+    default=False,
+    help='If specified, a RandomAgent or ModelAgent will act first, then the bot will take over')
+parser.add_option(
+    "--non_optimal_steps",
+    type=int,
+    default=None,
+    help='Number of non bot steps ModelAgent or RandomAgent takes before letting the bot take over'
+)
+parser.add_option(
+    "--model",
+    default=None,
+    help='Model to use to act for a few steps before letting the bot take over'
+)
+parser.add_option(
+    "--random_agent_seed",
+    type="int",
+    default=1,
+    help='Seed of the random agent that acts a few steps before letting the bot take over'
+)
+parser.add_option(
+    "--bad_action_proba",
+    type="float",
+    default=1.,
+    help='Probability of performing the non-optimal action when the random/model agent is performing'
+)
 parser.add_option(
     "--seed",
     type="int",
@@ -58,6 +98,15 @@
 if options.level:
     level_list = [options.level]
 
+bad_agent = None
+if options.advise_mode:
+    if options.model:
+        bad_agent = ModelAgent(options.model, obss_preprocessor=None,
+                               argmax=True)
+    else:
+        bad_agent = RandomAgent(seed=options.random_agent_seed,
+                                env=level_dict[level_list[0]]())
+
 start_time = time.time()
 
 for level_name in level_list:
@@ -76,10 +125,32 @@
         if options.verbose:
             print('%s/%s: %s, seed=%d' % (run_no+1, options.num_runs, mission.surface, mission_seed))
 
+        optimal_actions = []
+        before_optimal_actions = []
+        non_optimal_steps = options.non_optimal_steps or int(mission.max_steps // 3)
+        np.random.seed(mission_seed)
+
         try:
             episode_steps = 0
             while True:
                 action = expert.get_action()
+                if options.advise_mode and episode_steps < non_optimal_steps:
+                    if np.random.rand() < options.bad_action_proba:
+                        while True:
+                            action = bad_agent.act(mission.gen_obs())['action'].item()
+
+                            # To make things simple, only allow random left/right/fwd moves, and opening of doors
+                            if action in (mission.actions.left, mission.actions.right, mission.actions.forward):
+                                break
+                            fwd_pos = mission.agent_pos + mission.dir_vec
+                            fwd_cell = mission.grid.get(*fwd_pos)
+                            if action == mission.actions.toggle and fwd_cell is not None and fwd_cell.type == 'door':
+                                break
+
+                    before_optimal_actions.append(action)
+                else:
+                    optimal_actions.append(action)
+
                 expert.take_action(action)
                 obs, reward, done, info = mission.step(action)
 
@@ -91,11 +162,15 @@
                         num_success += 1
                         total_steps += episode_steps
                     if reward <= 0:
-                        print('FAILURE on %s, seed %d, reward %.2f' % (level_name, mission_seed, reward))
+                        assert episode_steps == mission.max_steps  # Is there another reason for this to happen ?
+                        if options.verbose:
+                            print('FAILURE on %s, seed %d, reward %.2f' % (level_name, mission_seed, reward))
                     break
         except Exception as e:
             print('FAILURE on %s, seed %d' % (level_name, mission_seed))
             print(e)
+            # Playing these 2 sets of actions should get you to the mission snapshot above
+            print(before_optimal_actions, optimal_actions)
 
     success_rate = 100 * num_success / options.num_runs
     mean_reward = total_reward / options.num_runs