Support subquests

scripts/tw-make

@@ -46,6 +46,8 @@ def parse_args():
                                help="Nb. of objects in the world.")
     custom_parser.add_argument("--quest-length", type=int, default=5, metavar="LENGTH",
                                help="Minimum nb. of actions the quest requires to be completed.")
+    custom_parser.add_argument("--quest-breadth", type=int, default=3, metavar="BREADTH",
+                               help="Control how non-linear a quest can be.")
 
     challenge_parser = subparsers.add_parser("challenge", parents=[general_parser],
                                              help='Generate a game for one of the challenges.')
@@ -72,7 +74,7 @@ if __name__ == "__main__":
     }
 
     if args.subcommand == "custom":
-        game_file, game = textworld.make(args.world_size, args.nb_objects, args.quest_length, grammar_flags,
+        game_file, game = textworld.make(args.world_size, args.nb_objects, args.quest_length, args.quest_breadth, grammar_flags,
                                          seed=args.seed, games_dir=args.output)
 
     elif args.subcommand == "challenge":
@@ -87,7 +89,7 @@ if __name__ == "__main__":
 
     print("Game generated: {}".format(game_file))
     if args.verbose:
-        print(game.quests[0].desc)
+        print(game.objective)
 
     if args.view:
         textworld.render.visualize(game, interactive=True)

scripts/tw-stats

@@ -38,7 +38,7 @@ if __name__ == "__main__":
             continue
 
         if len(game.quests) > 0:
-            objectives[game_filename] = game.quests[0].desc
+            objectives[game_filename] = game.objective
 
         names |= set(info.name for info in game.infos.values() if info.name is not None)
         game_logger.collect(game)

tests/test_make_game.py

@@ -11,21 +11,22 @@ def test_making_game_with_names_to_exclude():
     g_rng.set_seed(42)
 
     with make_temp_directory(prefix="test_render_wrapper") as tmpdir:
-        game_file1, game1 = textworld.make(2, 20, 3, {"names_to_exclude": []},
+        game_file1, game1 = textworld.make(2, 20, 3, 3, {"names_to_exclude": []},
                                            seed=123, games_dir=tmpdir)
 
         game1_objects_names = [info.name for info in game1.infos.values() if info.name is not None]
-        game_file2, game2 = textworld.make(2, 20, 3, {"names_to_exclude": game1_objects_names},
+        game_file2, game2 = textworld.make(2, 20, 3, 3, {"names_to_exclude": game1_objects_names},
                                            seed=123, games_dir=tmpdir)
         game2_objects_names = [info.name for info in game2.infos.values() if info.name is not None]
+        game2.grammar.flags.encode()
         assert len(set(game1_objects_names) & set(game2_objects_names)) == 0
 
 
 def test_making_game_is_reproducible_with_seed():
     grammar_flags = {}
     with make_temp_directory(prefix="test_render_wrapper") as tmpdir:
-        game_file1, game1 = textworld.make(2, 20, 3, grammar_flags, seed=123, games_dir=tmpdir)
-        game_file2, game2 = textworld.make(2, 20, 3, grammar_flags, seed=123, games_dir=tmpdir)
+        game_file1, game1 = textworld.make(2, 20, 3, 3, grammar_flags, seed=123, games_dir=tmpdir)
+        game_file2, game2 = textworld.make(2, 20, 3, 3, grammar_flags, seed=123, games_dir=tmpdir)
         assert game_file1 == game_file2
         assert game1 == game2
         # Make sure they are not the same Python objects.

tests/test_play_generated_games.py

@@ -16,12 +16,13 @@ def test_play_generated_games():
         # Sample game specs.
         world_size = rng.randint(1, 10)
         nb_objects = rng.randint(0, 20)
-        quest_length = rng.randint(1, 10)
+        quest_length = rng.randint(2, 5)
+        quest_breadth = rng.randint(3, 7)
         game_seed = rng.randint(0, 65365)
         grammar_flags = {}  # Default grammar.
 
         with make_temp_directory(prefix="test_play_generated_games") as tmpdir:
-            game_file, game = textworld.make(world_size, nb_objects, quest_length, grammar_flags,
+            game_file, game = textworld.make(world_size, nb_objects, quest_length, quest_breadth, grammar_flags,
                                              seed=game_seed, games_dir=tmpdir)
 
             # Solve the game using WalkthroughAgent.
@@ -47,10 +48,10 @@ def test_play_generated_games():
                 game_state, reward, done = env.step(command)
 
                 if done:
-                    msg = "Finished before playing `max_steps` steps."
+                    msg = "Finished before playing `max_steps` steps because of command '{}'.".format(command)
                     if game_state.has_won:
                         msg += " (winning)"
-                        assert game_state._game_progression.winning_policy == []
+                        assert len(game_state._game_progression.winning_policy) == 0
 
                     if game_state.has_lost:
                         msg += " (losing)"

tests/test_textworld.py

@@ -58,7 +58,7 @@ def test_game_walkthrough_agent(self):
         agent = textworld.agents.WalkthroughAgent()
         env = textworld.start(self.game_file)
         env.activate_state_tracking()
-        commands = self.game.quests[0].commands
+        commands = self.game.quests[-1].commands
         agent.reset(env)
         game_state = env.reset()
 

tests/test_tw_play.py → tests/test_tw-play.py

@@ -7,9 +7,9 @@
 from textworld.utils import make_temp_directory
 
 
-def test_making_a_custom_game():                                             
+def test_playing_a_game():                                             
     with make_temp_directory(prefix="test_tw-play") as tmpdir:    
-        game_file, _ = textworld.make(5, 10, 5, {}, seed=1234, games_dir=tmpdir)
+        game_file, _ = textworld.make(5, 10, 5, 4, {}, seed=1234, games_dir=tmpdir)
 
         command = ["tw-play", "--max-steps", "100", "--mode", "random", game_file]
         assert check_call(command) == 0

textworld/agents/walkthrough.py

@@ -26,7 +26,7 @@ def reset(self, env):
             raise NameError(msg)
 
         # Load command from the generated game.
-        self._commands = iter(env.game.quests[0].commands)
+        self._commands = iter(env.game.main_quest.commands)
 
     def act(self, game_state, reward, done):
         try:

textworld/envs/glulx/git_glulx_ml.py

@@ -146,13 +146,10 @@ def init(self, output: str, game=None,
         """
         output = _strip_input_prompt_symbol(output)
         super().init(output)
-        self._game_progression = GameProgression(game, track_quest=compute_intermediate_reward)
+        self._game_progression = GameProgression(game, track_quests=compute_intermediate_reward)
         self._state_tracking = state_tracking
         self._compute_intermediate_reward = compute_intermediate_reward and len(game.quests) > 0
-
-        self._objective = ""
-        if len(game.quests) > 0:
-            self._objective = game.quests[0].desc
+        self._objective = game.objective
 
     def view(self) -> "GlulxGameState":
         """
@@ -317,6 +314,7 @@ def intermediate_reward(self):
 
     @property
     def score(self):
+        # XXX: Should the score reflect the sum of all subquests' reward?
         if self.has_won:
             return 1
         elif self.has_lost:
@@ -326,6 +324,7 @@ def score(self):
 
     @property
     def max_score(self):
+        # XXX: Should the score reflect the sum of all subquests' reward?
         return 1
 
     @property

textworld/envs/wrappers/tests/test_viewer.py

@@ -17,7 +17,7 @@ def test_html_viewer():
     num_items = 10
     g_rng.set_seed(1234)
     grammar_flags = {"theme": "house", "include_adj": True}
-    game = textworld.generator.make_game(world_size=num_nodes, nb_objects=num_items, quest_length=3, grammar_flags=grammar_flags)
+    game = textworld.generator.make_game(world_size=num_nodes, nb_objects=num_items, quest_length=3, quest_breadth=1, grammar_flags=grammar_flags)
 
     game_name = "test_html_viewer_wrapper"
     with make_temp_directory(prefix=game_name) as tmpdir:

textworld/generator/__init__.py

@@ -147,7 +147,7 @@ def make_game_with(world, quests=None, grammar=None):
     return game
 
 
-def make_game(world_size: int, nb_objects: int, quest_length: int,
+def make_game(world_size: int, nb_objects: int, quest_length: int, quest_breadth: int,
               grammar_flags: Mapping = {},
               rngs: Optional[Dict[str, RandomState]] = None
               ) -> Game:
@@ -158,6 +158,7 @@ def make_game(world_size: int, nb_objects: int, quest_length: int,
         world_size: Number of rooms in the world.
         nb_objects: Number of objects in the world.
         quest_length: Minimum nb. of actions the quest requires to be completed.
+        quest_breadth: How many branches the quest can have.
         grammar_flags: Options for the grammar.
 
     Returns:
@@ -175,14 +176,34 @@ def make_game(world_size: int, nb_objects: int, quest_length: int,
     world = make_world(world_size, nb_objects=0, rngs=rngs)
 
     # Sample a quest according to quest_length.
-    options = ChainingOptions()
+    class Options(ChainingOptions):
+
+        def get_rules(self, depth):
+            if depth == 0:
+		        # Last action should not be "go <dir>".
+                return data.get_rules().get_matching("^(?!go.*).*")
+            else:
+                return super().get_rules(depth)
+
+    options = Options()
     options.backward = True
+    options.min_depth = 1
     options.max_depth = quest_length
+    options.min_breadth = 1
+    options.max_breadth = quest_breadth
     options.create_variables = True
     options.rng = rngs['rng_quest']
     options.restricted_types = {"r", "d"}
     chain = sample_quest(world.state, options)
+
+    subquests = []
+    for i in range(1, len(chain.nodes)):
+        if chain.nodes[i].breadth != chain.nodes[i - 1].breadth:
+            quest = Quest(chain.actions[:i])
+            subquests.append(quest)
+
     quest = Quest(chain.actions)
+    subquests.append(quest)
 
     # Set the initial state required for the quest.
     world.state = chain.initial_state
@@ -191,7 +212,9 @@ def make_game(world_size: int, nb_objects: int, quest_length: int,
     world.populate(nb_objects, rng=rngs['rng_objects'])
 
     grammar = make_grammar(grammar_flags, rng=rngs['rng_grammar'])
-    game = make_game_with(world, [quest], grammar)
+    game = make_game_with(world, subquests, grammar)
+    game.change_grammar(grammar)
+
     return game
 
 

textworld/generator/data/logic/door.twl

@@ -34,7 +34,13 @@ type d : t {
         link3 :: link(r, d, r') & link(r, d', r') -> fail();
 
         # There cannot be more than four doors in a room.
-        dr2 :: link(r, d1: d, r1: r) & link(r, d2: d, r2: r) & link(r, d3: d, r3: r) & link(r, d4: d, r4: r) & link(r, d5: d, r5: r) -> fail();
+        too_many_doors :: link(r, d1: d, r1: r) & link(r, d2: d, r2: r) & link(r, d3: d, r3: r) & link(r, d4: d, r4: r) & link(r, d5: d, r5: r) -> fail();
+
+        # There cannot be more than four doors in a room.
+        dr1 :: free(r, r1: r) & link(r, d2: d, r2: r) & link(r, d3: d, r3: r) & link(r, d4: d, r4: r) & link(r, d5: d, r5: r) -> fail();
+        dr2 :: free(r, r1: r) & free(r, r2: r) & link(r, d3: d, r3: r) & link(r, d4: d, r4: r) & link(r, d5: d, r5: r) -> fail();
+        dr3 :: free(r, r1: r) & free(r, r2: r) & free(r, r3: r) & link(r, d4: d, r4: r) & link(r, d5: d, r5: r) -> fail();
+        dr4 :: free(r, r1: r) & free(r, r2: r) & free(r, r3: r) & free(r, r4: r) & link(r, d5: d, r5: r) -> fail();
 
         free1 :: link(r, d, r') & free(r, r') & closed(d) -> fail();
         free2 :: link(r, d, r') & free(r, r') & locked(d) -> fail();

textworld/generator/data/text_grammars/house_instruction.twg

@@ -112,7 +112,7 @@ action_seperator_go/north: #afterhave# gone north, ;#emptyinstruction1#;#emptyin
 action_seperator_go/east: #afterhave# gone east, ;#emptyinstruction1#;#emptyinstruction2#;#emptyinstruction3#;#emptyinstruction4#;#emptyinstruction5#;#emptyinstruction6#;#emptyinstruction7#;#emptyinstruction8#;#emptyinstruction9#;#emptyinstruction10#
 action_seperator_go/west: #afterhave# gone west, ;#emptyinstruction1#;#emptyinstruction2#;#emptyinstruction3#;#emptyinstruction4#;#emptyinstruction5#;#emptyinstruction6#;#emptyinstruction7#;#emptyinstruction8#;#emptyinstruction9#;#emptyinstruction10#
 action_separator_close: #afterhave# #closed# the #close_open_types#, ; #after# #closing# the #close_open_types#, ;#emptyinstruction1#;#emptyinstruction2#;#emptyinstruction3#;#emptyinstruction4#;#emptyinstruction5#;#emptyinstruction6#;#emptyinstruction7#;#emptyinstruction8#;#emptyinstruction9#;#emptyinstruction10#
-action_separator_drop: #afterhave# #dropped# #obj_types#, ; #after# #dropping# #obj_types#, ;#emptyinstruction1#;#emptyinstruction2#;#emptyinstruction3#;#emptyinstruction4#;#emptyinstruction5#;#emptyinstruction6#;#emptyinstruction7#;#emptyinstruction8#;#emptyinstruction9#;#emptyinstruction10#
+action_separator_drop: #afterhave# #dropped# the #obj_types#, ; #after# #dropping# the #obj_types#, ;#emptyinstruction1#;#emptyinstruction2#;#emptyinstruction3#;#emptyinstruction4#;#emptyinstruction5#;#emptyinstruction6#;#emptyinstruction7#;#emptyinstruction8#;#emptyinstruction9#;#emptyinstruction10#
 #Separator Symbols
 afterhave:After you have;Having;Once you have;If you have
 havetaken:taken;got;picked up