bacow/easy-rl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update rainbowdqn

Browse Source
This commit is contained in:
johnjim0816
2022-05-31 01:20:58 +08:00
parent cfc0f6492e
commit c7c94468c9
149 changed files with 1866 additions and 1549 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
codes/envs/blackjack.py
View File

@@ -77,7 +77,7 @@ class BlackjackEnv(gym.Env):
self.natural = natural
# Start the first game
self._reset() # Number of
self.action_dim = 2
self.n_actions = 2
def reset(self):
return self._reset()

6
codes/envs/cliff_walking.py
View File

@@ -31,7 +31,7 @@ class CliffWalkingEnv(discrete.DiscreteEnv):
self.shape = (4, 12)
nS = np.prod(self.shape)
action_dim = 4
n_actions = 4
# Cliff Location
self._cliff = np.zeros(self.shape, dtype=np.bool)
@@ -41,7 +41,7 @@ class CliffWalkingEnv(discrete.DiscreteEnv):
P = {}
for s in range(nS):
position = np.unravel_index(s, self.shape)
P[s] = { a : [] for a in range(action_dim) }
P[s] = { a : [] for a in range(n_actions) }
P[s][UP] = self._calculate_transition_prob(position, [-1, 0])
P[s][RIGHT] = self._calculate_transition_prob(position, [0, 1])
P[s][DOWN] = self._calculate_transition_prob(position, [1, 0])
@@ -51,7 +51,7 @@ class CliffWalkingEnv(discrete.DiscreteEnv):
isd = np.zeros(nS)
isd[np.ravel_multi_index((3,0), self.shape)] = 1.0
super(CliffWalkingEnv, self).__init__(nS, action_dim, P, isd)
super(CliffWalkingEnv, self).__init__(nS, n_actions, P, isd)
def render(self, mode='human', close=False):
self._render(mode, close)

6
codes/envs/gridworld.py
View File

@@ -37,7 +37,7 @@ class GridworldEnv(discrete.DiscreteEnv):
self.shape = shape
nS = np.prod(shape)
action_dim = 4
n_actions = 4
MAX_Y = shape[0]
MAX_X = shape[1]
@@ -51,7 +51,7 @@ class GridworldEnv(discrete.DiscreteEnv):
y, x = it.multi_index
# P[s][a] = (prob, next_state, reward, is_done)
P[s] = {a : [] for a in range(action_dim)}
P[s] = {a : [] for a in range(n_actions)}
is_done = lambda s: s == 0 or s == (nS - 1)
reward = 0.0 if is_done(s) else -1.0
@@ -82,7 +82,7 @@ class GridworldEnv(discrete.DiscreteEnv):
# This should not be used in any model-free learning algorithm
self.P = P
super(GridworldEnv, self).__init__(nS, action_dim, P, isd)
super(GridworldEnv, self).__init__(nS, n_actions, P, isd)
def _render(self, mode='human', close=False):
""" Renders the current gridworld layout

195
codes/envs/gridworld_env.py
View File

@@ -1,195 +0,0 @@
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# -*- coding: utf-8 -*-
import gym
import turtle
import numpy as np
# turtle tutorial : https://docs.python.org/3.3/library/turtle.html
def GridWorld(gridmap=None, is_slippery=False):
if gridmap is None:
gridmap = ['SFFF', 'FHFH', 'FFFH', 'HFFG']
env = gym.make("FrozenLake-v0", desc=gridmap, is_slippery=False)
env = FrozenLakeWapper(env)
return env
class FrozenLakeWapper(gym.Wrapper):
def __init__(self, env):
gym.Wrapper.__init__(self, env)
self.max_y = env.desc.shape[0]
self.max_x = env.desc.shape[1]
self.t = None
self.unit = 50
def draw_box(self, x, y, fillcolor='', line_color='gray'):
self.t.up()
self.t.goto(x * self.unit, y * self.unit)
self.t.color(line_color)
self.t.fillcolor(fillcolor)
self.t.setheading(90)
self.t.down()
self.t.begin_fill()
for _ in range(4):
self.t.forward(self.unit)
self.t.right(90)
self.t.end_fill()
def move_player(self, x, y):
self.t.up()
self.t.setheading(90)
self.t.fillcolor('red')
self.t.goto((x + 0.5) * self.unit, (y + 0.5) * self.unit)
def render(self):
if self.t == None:
self.t = turtle.Turtle()
self.wn = turtle.Screen()
self.wn.setup(self.unit * self.max_x + 100,
self.unit * self.max_y + 100)
self.wn.setworldcoordinates(0, 0, self.unit * self.max_x,
self.unit * self.max_y)
self.t.shape('circle')
self.t.width(2)
self.t.speed(0)
self.t.color('gray')
for i in range(self.desc.shape[0]):
for j in range(self.desc.shape[1]):
x = j
y = self.max_y - 1 - i
if self.desc[i][j] == b'S': # Start
self.draw_box(x, y, 'white')
elif self.desc[i][j] == b'F': # Frozen ice
self.draw_box(x, y, 'white')
elif self.desc[i][j] == b'G': # Goal
self.draw_box(x, y, 'yellow')
elif self.desc[i][j] == b'H': # Hole
self.draw_box(x, y, 'black')
else:
self.draw_box(x, y, 'white')
self.t.shape('turtle')
x_pos = self.s % self.max_x
y_pos = self.max_y - 1 - int(self.s / self.max_x)
self.move_player(x_pos, y_pos)
class CliffWalkingWapper(gym.Wrapper):
def __init__(self, env):
gym.Wrapper.__init__(self, env)
self.t = None
self.unit = 50
self.max_x = 12
self.max_y = 4
def draw_x_line(self, y, x0, x1, color='gray'):
assert x1 > x0
self.t.color(color)
self.t.setheading(0)
self.t.up()
self.t.goto(x0, y)
self.t.down()
self.t.forward(x1 - x0)
def draw_y_line(self, x, y0, y1, color='gray'):
assert y1 > y0
self.t.color(color)
self.t.setheading(90)
self.t.up()
self.t.goto(x, y0)
self.t.down()
self.t.forward(y1 - y0)
def draw_box(self, x, y, fillcolor='', line_color='gray'):
self.t.up()
self.t.goto(x * self.unit, y * self.unit)
self.t.color(line_color)
self.t.fillcolor(fillcolor)
self.t.setheading(90)
self.t.down()
self.t.begin_fill()
for i in range(4):
self.t.forward(self.unit)
self.t.right(90)
self.t.end_fill()
def move_player(self, x, y):
self.t.up()
self.t.setheading(90)
self.t.fillcolor('red')
self.t.goto((x + 0.5) * self.unit, (y + 0.5) * self.unit)
def render(self):
if self.t == None:
self.t = turtle.Turtle()
self.wn = turtle.Screen()
self.wn.setup(self.unit * self.max_x + 100,
self.unit * self.max_y + 100)
self.wn.setworldcoordinates(0, 0, self.unit * self.max_x,
self.unit * self.max_y)
self.t.shape('circle')
self.t.width(2)
self.t.speed(0)
self.t.color('gray')
for _ in range(2):
self.t.forward(self.max_x * self.unit)
self.t.left(90)
self.t.forward(self.max_y * self.unit)
self.t.left(90)
for i in range(1, self.max_y):
self.draw_x_line(
y=i * self.unit, x0=0, x1=self.max_x * self.unit)
for i in range(1, self.max_x):
self.draw_y_line(
x=i * self.unit, y0=0, y1=self.max_y * self.unit)
for i in range(1, self.max_x - 1):
self.draw_box(i, 0, 'black')
self.draw_box(self.max_x - 1, 0, 'yellow')
self.t.shape('turtle')
x_pos = self.s % self.max_x
y_pos = self.max_y - 1 - int(self.s / self.max_x)
self.move_player(x_pos, y_pos)
if __name__ == '__main__':
# 环境1：FrozenLake, 可以配置冰面是否是滑的
# 0 left, 1 down, 2 right, 3 up
env = gym.make("FrozenLake-v0", is_slippery=False)
env = FrozenLakeWapper(env)
# 环境2：CliffWalking, 悬崖环境
# env = gym.make("CliffWalking-v0") # 0 up, 1 right, 2 down, 3 left
# env = CliffWalkingWapper(env)
# 环境3：自定义格子世界，可以配置地图, S为出发点Start, F为平地Floor, H为洞Hole, G为出口目标Goal
# gridmap = [
# 'SFFF',
# 'FHFF',
# 'FFFF',
# 'HFGF' ]
# env = GridWorld(gridmap)
env.reset()
for step in range(10):
action = np.random.randint(0, 4)
obs, reward, done, info = env.step(action)
print('step {}: action {}, obs {}, reward {}, done {}, info {}'.format(\
step, action, obs, reward, done, info))
# env.render() # 渲染一帧图像

BIN
codes/envs/snake/checkpoint.npy Normal file
View File

Binary file not shown.

BIN
codes/envs/snake/checkpoint1.npy Normal file
View File

Binary file not shown.

BIN
codes/envs/snake/checkpoint2.npy Normal file
View File

Binary file not shown.

BIN
codes/envs/snake/checkpoint3.npy Normal file
View File

Binary file not shown.

BIN
codes/envs/snake/q_agent.npy Normal file
View File

Binary file not shown.

12
codes/envs/stochastic_mdp.py
View File

@@ -17,31 +17,31 @@ class StochasticMDP:
def __init__(self):
self.end = False
self.curr_state = 2
self.action_dim = 2
self.state_dim = 6
self.n_actions = 2
self.n_states = 6
self.p_right = 0.5
def reset(self):
self.end = False
self.curr_state = 2
state = np.zeros(self.state_dim)
state = np.zeros(self.n_states)
state[self.curr_state - 1] = 1.
return state
def step(self, action):
if self.curr_state != 1:
if action == 1:
if random.random() < self.p_right and self.curr_state < self.state_dim:
if random.random() < self.p_right and self.curr_state < self.n_states:
self.curr_state += 1
else:
self.curr_state -= 1
if action == 0:
self.curr_state -= 1
if self.curr_state == self.state_dim:
if self.curr_state == self.n_states:
self.end = True
state = np.zeros(self.state_dim)
state = np.zeros(self.n_states)
state[self.curr_state - 1] = 1.
if self.curr_state == 1:

6
codes/envs/windy_gridworld.py
View File

@@ -30,7 +30,7 @@ class WindyGridworldEnv(discrete.DiscreteEnv):
self.shape = (7, 10)
nS = np.prod(self.shape)
action_dim = 4
n_actions = 4
# Wind strength
winds = np.zeros(self.shape)
@@ -41,7 +41,7 @@ class WindyGridworldEnv(discrete.DiscreteEnv):
P = {}
for s in range(nS):
position = np.unravel_index(s, self.shape)
P[s] = { a : [] for a in range(action_dim) }
P[s] = { a : [] for a in range(n_actions) }
P[s][UP] = self._calculate_transition_prob(position, [-1, 0], winds)
P[s][RIGHT] = self._calculate_transition_prob(position, [0, 1], winds)
P[s][DOWN] = self._calculate_transition_prob(position, [1, 0], winds)
@@ -51,7 +51,7 @@ class WindyGridworldEnv(discrete.DiscreteEnv):
isd = np.zeros(nS)
isd[np.ravel_multi_index((3,0), self.shape)] = 1.0
super(WindyGridworldEnv, self).__init__(nS, action_dim, P, isd)
super(WindyGridworldEnv, self).__init__(nS, n_actions, P, isd)
def render(self, mode='human', close=False):
self._render(mode, close)