manuel/ai-econ
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

its training good

Browse Source
This commit is contained in:
Manuel Plonski
2023-01-16 11:54:23 +01:00
parent 3a8d10e0b4
commit b7371502d3
3 changed files with 429 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

341
main working good econ trader univer.pys Normal file
View File

@@ -0,0 +1,341 @@
import numpy as np
from ai_economist import foundation
from stable_baselines3.common.vec_env import vec_frame_stack
from stable_baselines3.common.evaluation import evaluate_policy
from sb3_contrib.ppo_mask import MaskablePPO
import envs
import wrapper
import resources
from agents import trading_agent
from wrapper.base_econ_wrapper import BaseEconWrapper
from wrapper.reciever_econ_wrapper import RecieverEconWrapper
from wrapper.sb3_econ_converter import SB3EconConverter
from tqdm import tqdm
import components
from stable_baselines3.common.env_checker import check_env
from stable_baselines3 import PPO
from stable_baselines3.common.vec_env.vec_monitor import VecMonitor
from stable_baselines3.common.vec_env.vec_normalize import VecNormalize
from sb3_contrib import RecurrentPPO
from envs.econ_wrapper import EconVecEnv
from stable_baselines3.common.callbacks import BaseCallback
import yaml
import time
from threading import Thread
env_config = {
# ===== SCENARIO CLASS =====
# Which Scenario class to use: the class's name in the Scenario Registry (foundation.scenarios).
# The environment object will be an instance of the Scenario class.
'scenario_name': 'econ',
# ===== COMPONENTS =====
# Which components to use (specified as list of ("component_name", {component_kwargs}) tuples).
# "component_name" refers to the Component class's name in the Component Registry (foundation.components)
# {component_kwargs} is a dictionary of kwargs passed to the Component class
# The order in which components reset, step, and generate obs follows their listed order below.
'components': [
# (1) Building houses
('Craft', {'skill_dist': "pareto", 'commodities': ["Gem"],'max_skill_amount_benefit':1.5}),
# (2) Trading collectible resources
('ContinuousDoubleAuction', {'max_num_orders': 10}),
# (3) Movement and resource collection
('SimpleGather', {}),
('ExternalMarket',{'market_demand':{
'Gem': 15
}}),
],
# ===== SCENARIO CLASS ARGUMENTS =====
# (optional) kwargs that are added by the Scenario class (i.e. not defined in BaseEnvironment)
'starting_agent_coin': 10,
'fixed_four_skill_and_loc': True,
# ===== STANDARD ARGUMENTS ======
# kwargs that are used by every Scenario class (i.e. defined in BaseEnvironment)
'agent_composition': {"BasicMobileAgent": 20,"TradingAgent":5}, # Number of non-planner agents (must be > 1)
'world_size': [5, 5], # [Height, Width] of the env world
'episode_length': 256, # Number of timesteps per episode
'allow_observation_scaling': True,
'dense_log_frequency': 100,
'world_dense_log_frequency':1,
'energy_cost':0,
'energy_warmup_method': "auto",
'energy_warmup_constant': 4000,
# In multi-action-mode, the policy selects an action for each action subspace (defined in component code).
# Otherwise, the policy selects only 1 action.
'multi_action_mode_agents': False,
'multi_action_mode_planner': False,
# When flattening observations, concatenate scalar & vector observations before output.
# Otherwise, return observations with minimal processing.
'flatten_observations': False,
# When Flattening masks, concatenate each action subspace mask into a single array.
# Note: flatten_masks = True is required for masking action logits in the code below.
'flatten_masks': True,
}
eval_env_config = {
# ===== SCENARIO CLASS =====
# Which Scenario class to use: the class's name in the Scenario Registry (foundation.scenarios).
# The environment object will be an instance of the Scenario class.
'scenario_name': 'econ',
# ===== COMPONENTS =====
# Which components to use (specified as list of ("component_name", {component_kwargs}) tuples).
# "component_name" refers to the Component class's name in the Component Registry (foundation.components)
# {component_kwargs} is a dictionary of kwargs passed to the Component class
# The order in which components reset, step, and generate obs follows their listed order below.
'components': [
# (1) Building houses
('Craft', {'skill_dist': "pareto", 'commodities': ["Gem"],'max_skill_amount_benefit':1.5}),
# (2) Trading collectible resources
('ContinuousDoubleAuction', {'max_num_orders': 10}),
# (3) Movement and resource collection
('SimpleGather', {}),
('ExternalMarket',{'market_demand':{
'Gem': 15
}}),
],
# ===== SCENARIO CLASS ARGUMENTS =====
# (optional) kwargs that are added by the Scenario class (i.e. not defined in BaseEnvironment)
'starting_agent_coin': 10,
'fixed_four_skill_and_loc': True,
# ===== STANDARD ARGUMENTS ======
# kwargs that are used by every Scenario class (i.e. defined in BaseEnvironment)
'agent_composition': {"BasicMobileAgent": 20,"TradingAgent":5}, # Number of non-planner agents (must be > 1)
'world_size': [1, 1], # [Height, Width] of the env world
'episode_length': 256, # Number of timesteps per episode
'allow_observation_scaling': True,
'dense_log_frequency': 1,
'world_dense_log_frequency':1,
'energy_cost':0,
'energy_warmup_method': "auto",
'energy_warmup_constant': 4000,
# In multi-action-mode, the policy selects an action for each action subspace (defined in component code).
# Otherwise, the policy selects only 1 action.
'multi_action_mode_agents': False,
'multi_action_mode_planner': False,
# When flattening observations, concatenate scalar & vector observations before output.
# Otherwise, return observations with minimal processing.
'flatten_observations': False,
# When Flattening masks, concatenate each action subspace mask into a single array.
# Note: flatten_masks = True is required for masking action logits in the code below.
'flatten_masks': True,
}
num_frames=5
class TensorboardCallback(BaseCallback):
"""
Custom callback for plotting additional values in tensorboard.
"""
def __init__(self,econ, verbose=0):
super().__init__(verbose)
self.econ=econ
self.metrics=econ.scenario_metrics()
def _on_step(self) -> bool:
# Log scalar value (here a random variable)
prev_metrics=self.metrics
if self.econ.previous_episode_metrics is None:
self.metrics=self.econ.scenario_metrics()
else:
self.metrics=self.econ.previous_episode_metrics
curr_prod=self.metrics["social/productivity"]
trend_pord=curr_prod-prev_metrics["social/productivity"]
self.logger.record("social/total_productivity", curr_prod)
self.logger.record("social/delta_productivity", trend_pord)
return True
def printMarket(market):
for i in range(len(market)):
step=market[i]
if len(step)>0:
print("=== Step {} ===".format(i))
for transaction in step:
t=transaction
transstring = "({}) {} -> {} | [{}/{}] {} Coins\n".format(t["commodity"],t["seller"],t["buyer"],t["ask"],t["bid"],t["price"])
print(transstring)
return ""
def printBuilds(builds):
for i in range(len(builds)):
step=builds[i]
if len(step)>0:
for build in step:
t=build
transstring = "({}) Builder: {}, Skill: {}, Income {} ".format(i,t["builder"],t["build_skill"],t["income"])
print(transstring)
return ""
def printReplay(econ,agentid):
worldmaps=["Stone","Wood"]
log=econ.previous_episode_dense_log
agent=econ.world.agents[agentid]
agentid=str(agentid)
maxsetp=len(log["states"])-1
for step in range(maxsetp):
print()
print("=== Step {} ===".format(step))
# state
print("--- World ---")
world=log['world'][step]
for res in worldmaps:
print("{}: {}".format(res,world[res][0][0]))
print("--- State ---")
state=log['states'][step][agentid]
print(yaml.safe_dump(state))
print("--- Action ---")
action=log["actions"][step][agentid]
if action=={}:
print("Action: 0 -> NOOP")
else:
for k in action:
formats="Action: {}({})".format(k,action[k])
print(formats)
print("--- Reward ---")
reward=log["rewards"][step][agentid]
print("Reward: {}".format(reward))
#Setup Env Objects
econ=foundation.make_env_instance(**env_config)
market=econ.get_component("ContinuousDoubleAuction")
action=market.get_n_actions("TradingAgent")
baseEconWrapper=BaseEconWrapper(econ)
baseEconWrapper.run()
time.sleep(0.5)
mobileRecieverEconWrapper=RecieverEconWrapper(base_econ=baseEconWrapper,agent_classname="BasicMobileAgent")
tradeRecieverEconWrapper=RecieverEconWrapper(base_econ=baseEconWrapper,agent_classname="TradingAgent")
sb3_traderConverter=SB3EconConverter(tradeRecieverEconWrapper,econ,"TradingAgent",True)
sb3Converter=SB3EconConverter(mobileRecieverEconWrapper,econ,"BasicMobileAgent",True)
# attach sb3 wrappers
monenv=VecMonitor(venv=sb3Converter,info_keywords=["social/productivity","trend/productivity"])
montraidingenv=VecMonitor(venv=sb3_traderConverter)
stackenv_basic=vec_frame_stack.VecFrameStack(venv=monenv,n_stack=num_frames)
stackenv_traid=vec_frame_stack.VecFrameStack(venv=montraidingenv,n_stack=num_frames)
# Model setup complete
# Setup Eval Env
econ_eval=foundation.make_env_instance(**eval_env_config)
baseEconWrapper_eval=BaseEconWrapper(econ_eval)
baseEconWrapper_eval.run()
time.sleep(0.5)
mobileRecieverEconWrapper_eval=RecieverEconWrapper(base_econ=baseEconWrapper_eval,agent_classname="BasicMobileAgent")
tradeRecieverEconWrapper_eval=RecieverEconWrapper(base_econ=baseEconWrapper_eval,agent_classname="TradingAgent")
sb3_traderConverter_eval=SB3EconConverter(tradeRecieverEconWrapper_eval,econ_eval,"TradingAgent",False)
sb3Converter_eval=SB3EconConverter(mobileRecieverEconWrapper_eval,econ_eval,"BasicMobileAgent",False)
# attach sb3 wrappers
monenv_eval=VecMonitor(venv=sb3Converter_eval,info_keywords=["social/productivity","trend/productivity"])
montraidingenv_eval=VecMonitor(venv=sb3_traderConverter_eval)
stackenv_basic_eval=vec_frame_stack.VecFrameStack(venv=monenv_eval,n_stack=num_frames)
stackenv_traid_eval=vec_frame_stack.VecFrameStack(venv=montraidingenv_eval,n_stack=num_frames)
obs=monenv.reset()
# define training functions
def train(model,timesteps, econ_call,process_bar,name,db,index):
db[index]=model.learn(total_timesteps=timesteps,progress_bar=process_bar,reset_num_timesteps=False,tb_log_name=name,callback=TensorboardCallback(econ_call))
# prepare training
run_number=int(np.random.rand()*100)
runname="run_{}".format(run_number)
model_db=[None,None] # object for storing model
model = MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.99, learning_rate=1e-5,env=stackenv_basic, seed=300,verbose=1,device="cuda",tensorboard_log="./log")
model_trade=MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.99, learning_rate=1e-5,env=stackenv_traid, seed=300,verbose=1,device="cuda",tensorboard_log="./log")
n_agents=econ.n_agents
total_required_for_episode_basic=len(mobileRecieverEconWrapper.agnet_idx)*env_config['episode_length']
total_required_for_episode_traid=len(tradeRecieverEconWrapper.agnet_idx)*env_config['episode_length']
print("this is run {}".format(runname))
while True:
#Train
runname="run_{}_{}".format(run_number,"basic")
thread_model=Thread(target=train,args=(model,total_required_for_episode_basic*100,econ,True,runname,model_db,0))
runname="run_{}_{}".format(run_number,"trader")
thread_model_traid=Thread(target=train,args=(model_trade,total_required_for_episode_traid*100,econ,False,runname,model_db,1))
thread_model.start()
thread_model_traid.start()
thread_model.join()
thread_model_traid.join()
#normenv.save("temp-normalizer.ai")
## Run Eval
print("### EVAL ###")
obs_basic=stackenv_basic_eval.reset()
obs_trade=stackenv_traid_eval.reset()
model=model_db[0]
model_trade=model_db[1]
done=False
for i in tqdm(range(eval_env_config['episode_length'])):
#create masks
masks_basic=stackenv_basic_eval.action_masks()
masks_trade=stackenv_traid_eval.action_masks()
# get actions
action_basic=model.predict(obs_basic,action_masks=masks_basic)
action_trade=model_trade.predict(obs_trade,action_masks=masks_trade)
#submit async directly for non blocking operation
sb3Converter_eval.step_async(action_basic[0])
sb3_traderConverter_eval.step_async(action_trade[0])
# retieve full results
obs_basic,rew_basic,done_e,info=stackenv_basic_eval.step(action_basic[0])
obs_trade,rew_trade,done_e,info=stackenv_traid_eval.step(action_trade[0])
done=done_e[0]
market=econ_eval.get_component("ContinuousDoubleAuction")
craft=econ_eval.get_component("Craft")
# trades=market.get_dense_log()
build=craft.get_dense_log()
met=econ.previous_episode_metrics
printReplay(econ_eval,0)
# printMarket(trades)
# printBuilds(builds=build)
print("social/productivity: {}".format(met["social/productivity"]))
print("labor/weighted_cost: {}".format(met["labor/weighted_cost"]))
print("labor/warmup_integrator: {}".format(met["labor/warmup_integrator"]))
time.sleep(1)

121
main.py
View File

@@ -8,6 +8,7 @@ from sb3_contrib.ppo_mask import MaskablePPO
import envs
import wrapper
import resources
import pprint
from agents import trading_agent
from wrapper.base_econ_wrapper import BaseEconWrapper
from wrapper.reciever_econ_wrapper import RecieverEconWrapper
@@ -44,7 +45,7 @@ env_config = {
# (3) Movement and resource collection
('SimpleGather', {}),
('ExternalMarket',{'market_demand':{
'Gem': 10
'Gem': 15
}}),
],
@@ -99,7 +100,7 @@ eval_env_config = {
# (3) Movement and resource collection
('SimpleGather', {}),
('ExternalMarket',{'market_demand':{
'Gem': 10
'Gem': 15
}}),
],
@@ -115,7 +116,7 @@ eval_env_config = {
'world_size': [1, 1], # [Height, Width] of the env world
'episode_length': 256, # Number of timesteps per episode
'allow_observation_scaling': True,
'dense_log_frequency': 10,
'dense_log_frequency': 1,
'world_dense_log_frequency':1,
'energy_cost':0,
'energy_warmup_method': "auto",
@@ -134,7 +135,7 @@ eval_env_config = {
'flatten_masks': True,
}
num_frames=2
num_frames=5
class TensorboardCallback(BaseCallback):
"""
@@ -147,16 +148,17 @@ class TensorboardCallback(BaseCallback):
self.metrics=econ.scenario_metrics()
def _on_step(self) -> bool:
# Log scalar value (here a random variable)
prev_metrics=self.metrics
if self.econ.previous_episode_metrics is None:
self.metrics=self.econ.scenario_metrics()
else:
self.metrics=self.econ.previous_episode_metrics
curr_prod=self.metrics["social/productivity"]
trend_pord=curr_prod-prev_metrics["social/productivity"]
self.logger.record("social/total_productivity", curr_prod)
self.logger.record("social/delta_productivity", trend_pord)
if econ.world.timestep==0:
prev_metrics=self.metrics
if self.econ.previous_episode_metrics is None:
self.metrics=self.econ.scenario_metrics()
else:
self.metrics=self.econ.previous_episode_metrics
curr_prod=self.metrics["social/productivity"]
trend_pord=curr_prod-prev_metrics["social/productivity"]
self.logger.record("social/total_productivity", curr_prod)
self.logger.record("social/delta_productivity", trend_pord)
return True
@@ -200,7 +202,7 @@ def printReplay(econ,agentid):
print("--- State ---")
state=log['states'][step][agentid]
print(yaml.dump(state))
pprint.pprint(state)
print("--- Action ---")
action=log["actions"][step][agentid]
@@ -222,17 +224,40 @@ market=econ.get_component("ContinuousDoubleAuction")
action=market.get_n_actions("TradingAgent")
baseEconWrapper=BaseEconWrapper(econ)
baseEconWrapper.run()
time.sleep(0.5)
mobileRecieverEconWrapper=RecieverEconWrapper(base_econ=baseEconWrapper,agent_classname="BasicMobileAgent")
tradeRecieverEconWrapper=RecieverEconWrapper(base_econ=baseEconWrapper,agent_classname="TradingAgent")
sb3_traderConverter=SB3EconConverter(tradeRecieverEconWrapper,econ,"TradingAgent")
sb3Converter=SB3EconConverter(mobileRecieverEconWrapper,econ,"BasicMobileAgent")
#obs=sb3Converter.reset()
#vecenv=EconVecEnv(env_config=env_config)
sb3_traderConverter=SB3EconConverter(tradeRecieverEconWrapper,econ,"TradingAgent",True)
sb3Converter=SB3EconConverter(mobileRecieverEconWrapper,econ,"BasicMobileAgent",True)
# attach sb3 wrappers
monenv=VecMonitor(venv=sb3Converter,info_keywords=["social/productivity","trend/productivity"])
montraidingenv=VecMonitor(venv=sb3_traderConverter)
#normenv=VecNormalize(sb3Converter,norm_reward=False,clip_obs=1)
#stackenv=vec_frame_stack.VecFrameStack(venv=monenv,n_stack=10)
stackenv_basic=vec_frame_stack.VecFrameStack(venv=monenv,n_stack=num_frames)
stackenv_traid=vec_frame_stack.VecFrameStack(venv=montraidingenv,n_stack=num_frames)
# Model setup complete
# Setup Eval Env
econ_eval=foundation.make_env_instance(**eval_env_config)
baseEconWrapper_eval=BaseEconWrapper(econ_eval)
baseEconWrapper_eval.run()
time.sleep(0.5)
mobileRecieverEconWrapper_eval=RecieverEconWrapper(base_econ=baseEconWrapper_eval,agent_classname="BasicMobileAgent")
tradeRecieverEconWrapper_eval=RecieverEconWrapper(base_econ=baseEconWrapper_eval,agent_classname="TradingAgent")
sb3_traderConverter_eval=SB3EconConverter(tradeRecieverEconWrapper_eval,econ_eval,"TradingAgent",False)
sb3Converter_eval=SB3EconConverter(mobileRecieverEconWrapper_eval,econ_eval,"BasicMobileAgent",False)
# attach sb3 wrappers
monenv_eval=VecMonitor(venv=sb3Converter_eval,info_keywords=["social/productivity","trend/productivity"])
montraidingenv_eval=VecMonitor(venv=sb3_traderConverter_eval)
stackenv_basic_eval=vec_frame_stack.VecFrameStack(venv=monenv_eval,n_stack=num_frames)
stackenv_traid_eval=vec_frame_stack.VecFrameStack(venv=montraidingenv_eval,n_stack=num_frames)
obs=monenv.reset()
@@ -245,11 +270,11 @@ def train(model,timesteps, econ_call,process_bar,name,db,index):
# prepare training
run_number=int(np.random.rand()*100)
runname="run_{}".format(run_number)
model_db=[] # object for storing model
model_db=[None,None] # object for storing model
model = MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.98, learning_rate=5e-3,env=monenv, seed=225,verbose=1,device="cuda",tensorboard_log="./log")
model_trade=MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.98, learning_rate=5e-3,env=montraidingenv, seed=225,verbose=1,device="cuda",tensorboard_log="./log")
model = MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.99, learning_rate=1e-5,env=stackenv_basic, seed=300,verbose=1,device="cuda",tensorboard_log="./log")
model_trade=MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.99, learning_rate=1e-5,env=stackenv_traid, seed=300,verbose=1,device="cuda",tensorboard_log="./log")
n_agents=econ.n_agents
@@ -257,53 +282,57 @@ total_required_for_episode_basic=len(mobileRecieverEconWrapper.agnet_idx)*env_co
total_required_for_episode_traid=len(tradeRecieverEconWrapper.agnet_idx)*env_config['episode_length']
print("this is run {}".format(runname))
eval_econ=foundation.make_env_instance(**eval_env_config)
eval_base_econ=BaseEconWrapper(eval_econ)
eval_base_econ.run()
eval_mobileRecieverEconWrapper=RecieverEconWrapper(eval_base_econ,"BasicMobileAgent")
time.sleep(0.5)
eval_sb3_converter=SB3EconConverter(eval_mobileRecieverEconWrapper,eval_econ,"BasicMobileAgent")
while True:
# Create Eval ENV
vec_mon_eval=VecMonitor(venv=eval_sb3_converter)
#Train
runname="run_{}_{}".format(run_number,"basic")
thread_model=Thread(target=train,args=(model,total_required_for_episode_basic*300,econ,True,runname,model_db,0))
thread_model=Thread(target=train,args=(model,total_required_for_episode_basic*50,econ,True,runname,model_db,0))
runname="run_{}_{}".format(run_number,"trader")
thread_model_traid=Thread(target=train,args=(model_trade,total_required_for_episode_traid*300,econ,False,runname,model_db,1))
thread_model_traid=Thread(target=train,args=(model_trade,total_required_for_episode_traid*50,econ,False,runname,model_db,1))
thread_model.start()
thread_model_traid.start()
thread_model.join()
thread_model_traid.join()
#normenv.save("temp-normalizer.ai")
model=model_db[0]
model_trade=model_db[1]
model.save("basic.ai")
model_trade.save("trade.ai")
## Run Eval
print("### EVAL ###")
obs=vec_mon_eval.reset()
obs_basic=stackenv_basic_eval.reset()
obs_trade=stackenv_traid_eval.reset()
done=False
for i in tqdm(range(eval_env_config['episode_length'])):
masks=eval_sb3_converter.action_masks()
action=model.predict(obs,action_masks=masks)
obs,rew,done_e,info=vec_mon_eval.step(action[0])
#create masks
masks_basic=stackenv_basic_eval.action_masks()
masks_trade=stackenv_traid_eval.action_masks()
# get actions
action_basic=model.predict(obs_basic,action_masks=masks_basic)
action_trade=model_trade.predict(obs_trade,action_masks=masks_trade)
#submit async directly for non blocking operation
sb3Converter_eval.step_async(action_basic[0])
sb3_traderConverter_eval.step_async(action_trade[0])
# retieve full results
obs_basic,rew_basic,done_e,info=stackenv_basic_eval.step(action_basic[0])
obs_trade,rew_trade,done_e,info=stackenv_traid_eval.step(action_trade[0])
done=done_e[0]
market=eval_econ.get_component("ContinuousDoubleAuction")
craft=eval_econ.get_component("SimpleCraft")
market=econ_eval.get_component("ContinuousDoubleAuction")
craft=econ_eval.get_component("Craft")
# trades=market.get_dense_log()
build=craft.get_dense_log()
met=econ.previous_episode_metrics
printReplay(eval_econ,0)
printReplay(econ_eval,0)
# printMarket(trades)
printBuilds(builds=build)
# printBuilds(builds=build)
print("social/productivity: {}".format(met["social/productivity"]))
print("labor/weighted_cost: {}".format(met["labor/weighted_cost"]))
print("labor/warmup_integrator: {}".format(met["labor/warmup_integrator"]))

20
wrapper/sb3_econ_converter.py
View File

@@ -8,7 +8,7 @@ from typing import Any, Callable, List, Optional, Sequence, Type, Union
class SB3EconConverter(VecEnv, gym.Env):
def __init__(self, env: gym.Env, econ: base_env.BaseEnvironment,agentclass: str):
def __init__(self, env: gym.Env, econ: base_env.BaseEnvironment,agentclass: str,auto_reset: bool):
self.env=env
self.econ=econ
#get observation sample
@@ -20,7 +20,9 @@ class SB3EconConverter(VecEnv, gym.Env):
#flatten obervation of first agent
obs0=utils.package(obs[0],*self.packager)
obs0["flat"]
self.observation_space=gym.spaces.Box(low=-np.inf,high=np.inf,shape=(len(obs0["flat"]),),dtype=np.float32)
self.step_request_send=False
self.auto_reset=auto_reset
self.observation_space=gym.spaces.Box(low=0,high=np.inf,shape=(len(obs0["flat"]),),dtype=np.float32)
super().__init__(self.num_envs, self.observation_space, self.action_space)
@@ -30,9 +32,11 @@ class SB3EconConverter(VecEnv, gym.Env):
agent=self.econ.world.agents[idx]
return gym.spaces.Discrete(agent.action_spaces)
def step_async(self, actions: np.ndarray) -> None:
d_actions=utils.convert_gym_to_econ(actions)
return self.env.step_async(d_actions)
def step_async(self, actions: np.ndarray):
if self.step_request_send==False:
self.step_request_send=True
d_actions=utils.convert_gym_to_econ(actions)
return self.env.step_async(d_actions)
def step_wait(self) -> VecEnvStepReturn:
obs,rew,done,info=self.env.step_wait()
@@ -62,12 +66,14 @@ class SB3EconConverter(VecEnv, gym.Env):
for i in range(self.num_envs):
done_g[i]=done
c_info[i]["terminal_observation"]=c_obs[i]
c_obs=self.reset()
if self.auto_reset:
c_obs=self.reset()
self.step_request_send=False
return np.copy(c_obs),np.copy(c_rew),np.copy(done_g),np.copy(c_info)
def reset(self) -> VecEnvObs:
obs=self.env.reset()
self.step_request_send=False
f_obs={}
self.curr_obs=obs
for k,v in obs.items():