update bunch of miscellaneous things (#174)

* change RLBase.action_space to Base.OneTo(NUM_ACTIONS)

* rename AbstractGridWorldGame to AbstractGridWorld

* remove RLBaseEnvModule from benchmark.jl

* rename get_tile_map_height to get_height & similarly for width

* rename get_tile_pretty_repr & get_tile_map_pretty_repr to get_pretty_tile_map

* rename get_sub_tile_map_pretty_repr to get_pretty_sub_tile_map

* update printing for Catcher. Other envs broken temporarily

* update printing for CollectGems envs

* update miscellaneous methods for DoorKey envs

* update miscellaneous methods for DynamicObstacles envs

* update miscellaneous methods for GoToTarget envs

* update miscellaneous methods for GridRooms envs

* update miscellaneous methods for Maze envs

* update miscellaneous methods for SequentialRooms envs

* update miscellaneous methods for SingleRoom envs

* update miscellaneous methods for Snake

* update miscellaneous methods for Sokoban envs

* update miscellaneous methods for Transport envs

* update README

Author: Sid-Bhatia-0

README.md

@@ -4,7 +4,7 @@ A package for creating grid world environments for reinforcement learning in Jul
 
 This package is inspired by [gym-minigrid](https://github.com/maximecb/gym-minigrid). In order to cite this package, please refer to the file `CITATION.bib`. Starring the repository on GitHub is also appreciated. For benchmarks, refer to `benchmark/benchmarks.md`.
 
-**Important note:** This package is undergoing heavy internal redesign. This README reflects the new design. The README for the last released version (`0.4.0`) can be found [here](https://github.com/JuliaReinforcementLearning/GridWorlds.jl/tree/c0e86bb6c33819f0e4a4cefe0284d985d0474ed3).
+**Important note:** This package is undergoing heavy internal redesign. This README reflects the new design (some visualizations might be oudated). The README for the last released version (`0.4.0`) can be found [here](https://github.com/JuliaReinforcementLearning/GridWorlds.jl/tree/c0e86bb6c33819f0e4a4cefe0284d985d0474ed3).
 
 ## Table of contents:
 

benchmark/benchmark.jl

@@ -22,7 +22,7 @@ end
 function benchmark(Env, num_resets, steps_per_reset)
     benchmark = DS.OrderedDict()
 
-    env = GW.RLBaseEnvModule.RLBaseEnv(Env())
+    env = GW.RLBaseEnv(Env())
 
     benchmark[:random_policy] = BT.@benchmark run_random_policy!($(Ref(env))[], $(Ref(num_resets))[], $(Ref(steps_per_reset))[])
     benchmark[:reset] = BT.@benchmark RLBase.reset!($(Ref(env))[])

src/abstract_grid_world.jl

@@ -1,57 +1,58 @@
-abstract type AbstractGridWorldGame end
+abstract type AbstractGridWorld end
 
 #####
 ##### Game logic methods
 #####
 
-reset!(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
-act!(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
+reset!(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
+act!(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
 
 #####
 ##### Optional methods for pretty printing, playing, etc...
 #####
 
-get_tile_pretty_repr(env::AbstractGridWorldGame, i::Integer, j::Integer) = error("Method not implemented for $(typeof(env))")
-get_sub_tile_map_pretty_repr(env::AbstractGridWorldGame, position::CartesianIndex{2}) = error("Method not implemented for $(typeof(env))")
-get_action_keys(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
-get_action_names(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
-get_tile_map_height(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
-get_tile_map_width(env::AbstractGridWorldGame) = error("Method not implemented for $(typeof(env))")
+get_pretty_tile_map(env::AbstractGridWorld, i::Integer, j::Integer) = error("Method not implemented for $(typeof(env))")
+get_pretty_sub_tile_map(env::AbstractGridWorld, position::CartesianIndex{2}) = error("Method not implemented for $(typeof(env))")
+get_action_keys(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
+get_action_names(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
+get_object_names(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
+get_height(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
+get_width(env::AbstractGridWorld) = error("Method not implemented for $(typeof(env))")
 
-function get_tile_map_pretty_repr(env::AbstractGridWorldGame)
-    height_tile_map = get_tile_map_height(env)
-    width_tile_map = get_tile_map_width(env)
+function get_pretty_tile_map(env::AbstractGridWorld)
+    height = get_height(env)
+    width = get_width(env)
 
     str = ""
 
-    for i in 1:height_tile_map
-        for j in 1:width_tile_map
-            str = str * get_tile_pretty_repr(env, i, j)
+    for i in 1:height
+        for j in 1:width
+            str = str * get_pretty_tile_map(env, CartesianIndex(i, j))
         end
-        if i < height_tile_map
+        if i < height
             str = str * "\n"
         end
     end
 
     return str
 end
 
-function get_window_size(env::AbstractGridWorldGame)
-    height = get_tile_map_height(env)
-    width = get_tile_map_width(env)
+function get_window_size(env::AbstractGridWorld)
+    height = get_height(env)
+    width = get_width(env)
     return (2 * (height ÷ 4) + 1, 2 * (width ÷ 4) + 1)
 end
 
-function get_sub_tile_map_pretty_repr(env::AbstractGridWorldGame, window_size)
-    height_sub_tile_map, width_sub_tile_map = window_size
+function get_pretty_sub_tile_map(env::AbstractGridWorld, window_size)
+    height, width = window_size
 
     str = ""
 
-    for i in 1:height_sub_tile_map
-        for j in 1:width_sub_tile_map
-            str = str * get_sub_tile_map_pretty_repr(env, window_size, CartesianIndex(i, j))
+    for i in 1:height
+        for j in 1:width
+            str = str * get_pretty_sub_tile_map(env, window_size, CartesianIndex(i, j))
         end
-        if i < height_sub_tile_map
+        if i < height
             str = str * "\n"
         end
     end

src/envs/catcher.jl

@@ -13,7 +13,7 @@ const AGENT = 1
 const GEM = 2
 const NUM_ACTIONS = 3
 
-mutable struct Catcher{R, RNG} <: GW.AbstractGridWorldGame
+mutable struct Catcher{R, RNG} <: GW.AbstractGridWorld
     tile_map::BitArray{3}
     agent_position::CartesianIndex{2}
     reward::R
@@ -125,30 +125,35 @@ end
 ##### miscellaneous
 #####
 
-CHARACTERS = ('☻', '♦', '⋅')
+GW.get_height(env::Catcher) = size(env.tile_map, 2)
+GW.get_width(env::Catcher) = size(env.tile_map, 3)
 
-GW.get_tile_map_height(env::Catcher) = size(env.tile_map, 2)
-GW.get_tile_map_width(env::Catcher) = size(env.tile_map, 3)
+function GW.get_pretty_tile_map(env::Catcher, position::CartesianIndex{2})
+    characters = ('☻', '♦', '⋅')
 
-function GW.get_tile_pretty_repr(env::Catcher, i::Integer, j::Integer)
-    object = findfirst(@view env.tile_map[:, i, j])
+    object = findfirst(@view env.tile_map[:, position])
     if isnothing(object)
-        return CHARACTERS[end]
+        return characters[end]
     else
-        return CHARACTERS[object]
+        return characters[object]
     end
 end
 
-GW.get_action_keys(env::Catcher) = ('a', 'd', 's')
+GW.get_object_names(env::Catcher) = (:AGENT, :GEM)
 GW.get_action_names(env::Catcher) = (:MOVE_LEFT, :MOVE_RIGHT, :NO_MOVE)
 
 function Base.show(io::IO, ::MIME"text/plain", env::Catcher)
-    str = GW.get_tile_map_pretty_repr(env)
-    str = str * "\nreward = $(env.reward)\ndone = $(env.done)"
+    str = "tile_map:\n"
+    str = str * GW.get_pretty_tile_map(env)
+    str = str * "\nreward: $(env.reward)\ndone: $(env.done)"
+    str = str * "\naction_names: $(GW.get_action_names(env))"
+    str = str * "\nobject_names: $(GW.get_object_names(env))"
     print(io, str)
     return nothing
 end
 
+GW.get_action_keys(env::Catcher) = ('a', 'd', 's')
+
 #####
 ##### RLBase API
 #####
@@ -159,7 +164,7 @@ RLBase.state(env::GW.RLBaseEnv{E}, ::RLBase.InternalState) where {E <: Catcher}
 
 RLBase.reset!(env::GW.RLBaseEnv{E}) where {E <: Catcher} = GW.reset!(env.env)
 
-RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: Catcher} = 1:NUM_ACTIONS
+RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: Catcher} = Base.OneTo(NUM_ACTIONS)
 (env::GW.RLBaseEnv{E})(action) where {E <: Catcher} = GW.act!(env.env, action)
 
 RLBase.reward(env::GW.RLBaseEnv{E}) where {E <: Catcher} = env.env.reward

src/envs/collect_gems_directed.jl

@@ -15,7 +15,7 @@ const WALL = CGUM.WALL
 const GEM = CGUM.GEM
 const NUM_ACTIONS = 4
 
-mutable struct CollectGemsDirected{R, RNG} <: GW.AbstractGridWorldGame
+mutable struct CollectGemsDirected{R, RNG} <: GW.AbstractGridWorld
     env::CGUM.CollectGemsUndirected{R, RNG}
     agent_direction::Int
 end
@@ -78,32 +78,58 @@ end
 ##### miscellaneous
 #####
 
-CHARACTERS = ('☻', '█', '♦', '→', '↑', '←', '↓', '⋅')
+GW.get_height(env::CollectGemsDirected) = GW.get_height(env.env)
+GW.get_width(env::CollectGemsDirected) = GW.get_width(env.env)
 
-GW.get_tile_map_height(env::CollectGemsDirected) = size(env.env.tile_map, 2)
-GW.get_tile_map_width(env::CollectGemsDirected) = size(env.env.tile_map, 3)
+GW.get_action_names(env::CollectGemsDirected) = (:MOVE_FORWARD, :MOVE_BACKWARD, :TURN_LEFT, :TURN_RIGHT)
+GW.get_object_names(env::CollectGemsDirected) = GW.get_object_names(env.env)
+
+function GW.get_pretty_tile_map(env::CollectGemsDirected, position::CartesianIndex{2})
+    characters = ('☻', '█', '♦', '→', '↑', '←', '↓', '⋅')
 
-function GW.get_tile_pretty_repr(env::CollectGemsDirected, i::Integer, j::Integer)
-    object = findfirst(@view env.env.tile_map[:, i, j])
+    object = findfirst(@view env.env.tile_map[:, position])
     if isnothing(object)
-        return CHARACTERS[end]
+        return characters[end]
     elseif object == AGENT
-        return CHARACTERS[NUM_OBJECTS + 1 + env.agent_direction]
+        return characters[NUM_OBJECTS + 1 + env.agent_direction]
     else
-        return CHARACTERS[object]
+        return characters[object]
     end
 end
 
-GW.get_action_keys(env::CollectGemsDirected) = ('w', 's', 'a', 'd')
-GW.get_action_names(env::CollectGemsDirected) = (:MOVE_FORWARD, :MOVE_BACKWARD, :TURN_LEFT, :TURN_RIGHT)
+function GW.get_pretty_sub_tile_map(env::CollectGemsDirected, window_size, position::CartesianIndex{2})
+    tile_map = env.env.tile_map
+    agent_position = env.env.agent_position
+    agent_direction = env.agent_direction
+
+    characters = ('☻', '█', '♦', '→', '↑', '←', '↓', '⋅')
+
+    sub_tile_map = GW.get_sub_tile_map(tile_map, agent_position, window_size, agent_direction)
+
+    object = findfirst(@view sub_tile_map[:, position])
+    if isnothing(object)
+        return characters[end]
+    elseif object == AGENT
+        return '↓'
+    else
+        return characters[object]
+    end
+end
 
 function Base.show(io::IO, ::MIME"text/plain", env::CollectGemsDirected)
-    str = GW.get_tile_map_pretty_repr(env)
-    str = str * "\nreward = $(env.env.reward)\ndone = $(env.env.done)"
+    str = "tile_map:\n"
+    str = str * GW.get_pretty_tile_map(env)
+    str = str * "\nsub_tile_map:\n"
+    str = str * GW.get_pretty_sub_tile_map(env, GW.get_window_size(env))
+    str = str * "\nreward: $(env.env.reward)\ndone: $(env.env.done)"
+    str = str * "\naction_names: $(GW.get_action_names(env))"
+    str = str * "\nobject_names: $(GW.get_object_names(env))"
     print(io, str)
     return nothing
 end
 
+GW.get_action_keys(env::CollectGemsDirected) = ('w', 's', 'a', 'd')
+
 #####
 ##### RLBase API
 #####
@@ -114,7 +140,7 @@ RLBase.state(env::GW.RLBaseEnv{E}, ::RLBase.InternalState) where {E <: CollectGe
 
 RLBase.reset!(env::GW.RLBaseEnv{E}) where {E <: CollectGemsDirected} = GW.reset!(env.env)
 
-RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: CollectGemsDirected} = 1:NUM_ACTIONS
+RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: CollectGemsDirected} = Base.OneTo(NUM_ACTIONS)
 (env::GW.RLBaseEnv{E})(action) where {E <: CollectGemsDirected} = GW.act!(env.env, action)
 
 RLBase.reward(env::GW.RLBaseEnv{E}) where {E <: CollectGemsDirected} = env.env.env.reward

src/envs/collect_gems_multi_agent_undirected.jl

@@ -10,7 +10,7 @@ import ReinforcementLearningBase as RLBase
 
 const NUM_ACTIONS = 4
 
-mutable struct CollectGemsMultiAgentUndirected{R, RNG} <: GW.AbstractGridWorldGame
+mutable struct CollectGemsMultiAgentUndirected{R, RNG} <: GW.AbstractGridWorld
     tile_map::BitArray{3}
     agent_positions::Vector{CartesianIndex{2}}
     current_agent::Int
@@ -97,7 +97,7 @@ function GW.reset!(env::CollectGemsMultiAgentUndirected)
 end
 
 function GW.act!(env::CollectGemsMultiAgentUndirected, action)
-    @assert action in 1:NUM_ACTIONS "Invalid action $(action)"
+    @assert action in Base.OneTo(NUM_ACTIONS) "Invalid action $(action)"
 
     tile_map = env.tile_map
     agent_positions = env.agent_positions
@@ -148,12 +148,26 @@ end
 ##### miscellaneous
 #####
 
-GW.get_tile_map_height(env::CollectGemsMultiAgentUndirected) = size(env.tile_map, 2)
-GW.get_tile_map_width(env::CollectGemsMultiAgentUndirected) = size(env.tile_map, 3)
+GW.get_height(env::CollectGemsMultiAgentUndirected) = size(env.tile_map, 2)
+GW.get_width(env::CollectGemsMultiAgentUndirected) = size(env.tile_map, 3)
 
-function GW.get_tile_pretty_repr(env::CollectGemsMultiAgentUndirected, i::Integer, j::Integer)
+GW.get_action_names(env::CollectGemsMultiAgentUndirected) = (:MOVE_UP, :MOVE_DOWN, :MOVE_LEFT, :MOVE_RIGHT)
+
+function GW.get_object_names(env::CollectGemsMultiAgentUndirected)
+    num_agents = length(env.agent_positions)
+    object_names = Array{Symbol}(undef, num_agents + 2)
+    for i in 1:num_agents
+        object_names[i] = Symbol("AGENT", "$(i)")
+    end
+    object_names[end - 1] = :WALL
+    object_names[end] = :GEM
+
+    return object_names
+end
+
+function GW.get_pretty_tile_map(env::CollectGemsMultiAgentUndirected, position::CartesianIndex{2})
     tile_map = env.tile_map
-    object = findfirst(@view tile_map[:, i, j])
+    object = findfirst(@view tile_map[:, position])
     num_agents = size(tile_map, 1) - 2
 
     if isnothing(object)
@@ -167,16 +181,40 @@ function GW.get_tile_pretty_repr(env::CollectGemsMultiAgentUndirected, i::Intege
     end
 end
 
-GW.get_action_keys(env::CollectGemsMultiAgentUndirected) = ('w', 's', 'a', 'd')
-GW.get_action_names(env::CollectGemsMultiAgentUndirected) = (:MOVE_UP, :MOVE_DOWN, :MOVE_LEFT, :MOVE_RIGHT)
+function GW.get_pretty_sub_tile_map(env::CollectGemsMultiAgentUndirected, window_size, position::CartesianIndex{2})
+    tile_map = env.tile_map
+    agent_positions = env.agent_positions
+    agent_position = agent_positions[env.current_agent]
+    num_agents = length(agent_positions)
+
+    sub_tile_map = GW.get_sub_tile_map(tile_map, agent_position, window_size)
+
+    object = findfirst(@view sub_tile_map[:, position])
+    if isnothing(object)
+        return "⋅"
+    elseif object in 1 : num_agents
+        return "$(object)"
+    elseif object == num_agents + 1
+        return "█"
+    else
+        return "♦"
+    end
+end
 
 function Base.show(io::IO, ::MIME"text/plain", env::CollectGemsMultiAgentUndirected)
-    str = GW.get_tile_map_pretty_repr(env)
-    str = str * "\nreward = $(env.reward)\ndone = $(env.done)\ncurrent_agent = $(env.current_agent)"
+    str = "tile_map:\n"
+    str = str * GW.get_pretty_tile_map(env)
+    str = str * "\nsub_tile_map:\n"
+    str = str * GW.get_pretty_sub_tile_map(env, GW.get_window_size(env))
+    str = str * "\nreward: $(env.reward)\ndone: $(env.done)\ncurrent_agent = $(env.current_agent)"
+    str = str * "\naction_names: $(GW.get_action_names(env))"
+    str = str * "\nobject_names: $(GW.get_object_names(env))"
     print(io, str)
     return nothing
 end
 
+GW.get_action_keys(env::CollectGemsMultiAgentUndirected) = ('w', 's', 'a', 'd')
+
 #####
 ##### RLBase API
 #####
@@ -187,7 +225,7 @@ RLBase.state(env::GW.RLBaseEnv{E}, ::RLBase.InternalState) where {E <: CollectGe
 
 RLBase.reset!(env::GW.RLBaseEnv{E}) where {E <: CollectGemsMultiAgentUndirected} = GW.reset!(env.env)
 
-RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: CollectGemsMultiAgentUndirected} = 1:NUM_ACTIONS
+RLBase.action_space(env::GW.RLBaseEnv{E}) where {E <: CollectGemsMultiAgentUndirected} = Base.OneTo(NUM_ACTIONS)
 (env::GW.RLBaseEnv{E})(action) where {E <: CollectGemsMultiAgentUndirected} = GW.act!(env.env, action)
 
 RLBase.reward(env::GW.RLBaseEnv{E}) where {E <: CollectGemsMultiAgentUndirected} = env.env.reward