Compare commits
5
Commits
731aad0a7b
...
ui
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
a137f37603 | ||
|
|
8c36302edc | ||
|
|
f1982d1fe4 | ||
|
|
1ee1c2eb04 | ||
|
|
bfd6d1e37b |
+3
-1
@@ -1,4 +1,6 @@
|
||||
logs/*
|
||||
_pycache_
|
||||
*.pyc
|
||||
*tfevents*
|
||||
*tfevents*
|
||||
db/econ_data/*
|
||||
db/metabase_data/*
|
||||
Binary file not shown.
@@ -0,0 +1,42 @@
|
||||
version: "3.5"
|
||||
|
||||
services:
|
||||
|
||||
postgres:
|
||||
image: postgres:10
|
||||
volumes:
|
||||
- "./metabase_data:/var/lib/postgresql/data"
|
||||
restart: always
|
||||
environment:
|
||||
POSTGRES_DB: metabase
|
||||
POSTGRES_USER: metabase_usr
|
||||
POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
|
||||
|
||||
metabase:
|
||||
depends_on:
|
||||
- postgres
|
||||
image: metabase/metabase
|
||||
ports:
|
||||
- "3000:3000"
|
||||
restart: always
|
||||
environment:
|
||||
MB_ENCRYPTION_SECRET_KEY: ${MB}
|
||||
MB_DB_TYPE: postgres
|
||||
MB_DB_DBNAME: metabase
|
||||
MB_DB_PORT: 5432
|
||||
MB_DB_USER: metabase_usr
|
||||
MB_DB_PASS: ${POSTGRES_PASSWORD}
|
||||
MB_DB_HOST: postgres
|
||||
|
||||
mongo:
|
||||
image: mongo
|
||||
restart: always
|
||||
environment:
|
||||
MONGO_INITDB_DATABASE: econ
|
||||
MONGO_INITDB_ROOT_USERNAME: root
|
||||
MONGO_INITDB_ROOT_PASSWORD: econ
|
||||
ports:
|
||||
- '27017:27017'
|
||||
volumes:
|
||||
- ./econ_data:/data/db
|
||||
|
||||
@@ -0,0 +1,58 @@
|
||||
from pymongo import MongoClient
|
||||
import datetime
|
||||
cmt="current_market_trades"
|
||||
cma="current_market_asks"
|
||||
cmb="current_market_bids"
|
||||
|
||||
class MongoExporter():
|
||||
|
||||
def __init__(self,conn_string,db,econ) -> None:
|
||||
|
||||
# Provide the mongodb atlas url to connect python to mongodb using pymongo
|
||||
|
||||
|
||||
# Create a connection using MongoClient. You can import MongoClient or use pymongo.MongoClient
|
||||
client = MongoClient(conn_string)
|
||||
|
||||
# Create the database for our example (we will use the same database throughout the tutorial
|
||||
self.client=client[db]
|
||||
self.econ=econ
|
||||
|
||||
def reset_market(self):
|
||||
"""Resets the current market data in the analytics stack"""
|
||||
c=self.client
|
||||
c[cmt].drop()
|
||||
c[cma].drop()
|
||||
c[cmb].drop()
|
||||
|
||||
def submit_full_market(self):
|
||||
c=self.client
|
||||
market=self.econ.get_component("ContinuousDoubleAuction")
|
||||
self.reset_market()
|
||||
base=datetime.datetime.fromtimestamp(0)
|
||||
delta=datetime.timedelta(days=1)
|
||||
for i in range(len(market.executed_trades)):
|
||||
step=market.executed_trades[i]
|
||||
timestep=base+(i*delta)
|
||||
if len(step)>0:
|
||||
for transaction in step:
|
||||
transaction["step"]=i
|
||||
c[cmt].insert_one(transaction)
|
||||
for com,asks in market.asks.items():
|
||||
for ask in asks:
|
||||
ask["commodity"]=com
|
||||
c[cma].insert_one(ask)
|
||||
for com,bids in market.bids.items():
|
||||
for bid in bids:
|
||||
bid["commodity"]=com
|
||||
c[cmb].insert_one(bid)
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
# This is added so that many files can reuse the function get_database()
|
||||
@@ -0,0 +1,362 @@
|
||||
|
||||
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
|
||||
import pprint
|
||||
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': 50,
|
||||
'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
|
||||
'isoelastic_eta':0.001,
|
||||
'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': 50,
|
||||
'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,
|
||||
'isoelastic_eta':0.001,
|
||||
'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=1
|
||||
|
||||
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)
|
||||
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/np.maximum(0.0001,prev_metrics["social/productivity"])
|
||||
self.logger.record("social/total_productivity", curr_prod)
|
||||
self.logger.record("social/delta_productivity", trend_pord)
|
||||
|
||||
return True
|
||||
|
||||
min_at_target_basic=0.5
|
||||
min_lr_basic=5e-6
|
||||
start_lr_basic=9e-4
|
||||
|
||||
min_at_target_trade=0.5
|
||||
min_lr_trade=5e-6
|
||||
start_lr_trade=9e-4
|
||||
|
||||
def learning_rate_adj_basic(x) -> float:
|
||||
diff=start_lr_basic-min_lr_basic
|
||||
lr=min_lr_basic+x*diff
|
||||
return lr
|
||||
|
||||
def learning_rate_adj_trade(x) -> float:
|
||||
diff=start_lr_trade-min_lr_trade
|
||||
lr=min_lr_basic+x*diff
|
||||
return lr
|
||||
|
||||
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]
|
||||
|
||||
pprint.pprint(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=0.001, vf_coef=0.5 ,gamma=0.99, learning_rate=learning_rate_adj_basic,env=stackenv_basic, seed=445,verbose=1,device="cuda",tensorboard_log="./log")
|
||||
model_trade=MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.001, vf_coef=0.5 ,gamma=0.99, learning_rate=learning_rate_adj_trade,env=stackenv_traid, seed=445,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*150,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*150,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_basic=stackenv_basic_eval.reset()
|
||||
obs_trade=stackenv_traid_eval.reset()
|
||||
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)
|
||||
|
||||
|
||||
|
||||
@@ -41,11 +41,11 @@ env_config = {
|
||||
# (1) Building houses
|
||||
('Craft', {'skill_dist': "pareto", 'commodities': ["Gem"],'max_skill_amount_benefit':1.5}),
|
||||
# (2) Trading collectible resources
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10}),
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10,'max_bid_ask':100}),
|
||||
# (3) Movement and resource collection
|
||||
('SimpleGather', {}),
|
||||
('ExternalMarket',{'market_demand':{
|
||||
'Gem': 15
|
||||
'Gem': 65
|
||||
}}),
|
||||
],
|
||||
|
||||
@@ -57,14 +57,14 @@ env_config = {
|
||||
|
||||
# ===== 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)
|
||||
'agent_composition': {"BasicMobileAgent": 10,"TradingAgent":2}, # 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
|
||||
'isoelastic_eta':0.001,
|
||||
'allow_observation_scaling': True,
|
||||
'dense_log_frequency': 100,
|
||||
'world_dense_log_frequency':1,
|
||||
'energy_cost':0,
|
||||
'energy_cost':0.2,
|
||||
'energy_warmup_method': "auto",
|
||||
'energy_warmup_constant': 4000,
|
||||
|
||||
@@ -97,7 +97,7 @@ eval_env_config = {
|
||||
# (1) Building houses
|
||||
('Craft', {'skill_dist': "pareto", 'commodities': ["Gem"],'max_skill_amount_benefit':1.5}),
|
||||
# (2) Trading collectible resources
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10}),
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10,'max_bid_ask':100}),
|
||||
# (3) Movement and resource collection
|
||||
('SimpleGather', {}),
|
||||
('ExternalMarket',{'market_demand':{
|
||||
@@ -113,14 +113,14 @@ eval_env_config = {
|
||||
|
||||
# ===== 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)
|
||||
'agent_composition': {"BasicMobileAgent": 10,"TradingAgent":10}, # 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,
|
||||
'isoelastic_eta':0.001,
|
||||
'dense_log_frequency': 1,
|
||||
'world_dense_log_frequency':1,
|
||||
'energy_cost':0,
|
||||
'energy_cost':0.2,
|
||||
'energy_warmup_method': "auto",
|
||||
'energy_warmup_constant': 4000,
|
||||
|
||||
@@ -138,6 +138,7 @@ eval_env_config = {
|
||||
}
|
||||
|
||||
num_frames=1
|
||||
CONNECTION_STRING = "mongodb+srv://root:econ@localhost:/econ"
|
||||
|
||||
class TensorboardCallback(BaseCallback):
|
||||
"""
|
||||
@@ -157,7 +158,7 @@ class TensorboardCallback(BaseCallback):
|
||||
else:
|
||||
self.metrics=self.econ.previous_episode_metrics
|
||||
curr_prod=self.metrics["social/productivity"]
|
||||
trend_pord=curr_prod-prev_metrics["social/productivity"]
|
||||
trend_pord=curr_prod/np.maximum(0.0001,prev_metrics["social/productivity"])
|
||||
self.logger.record("social/total_productivity", curr_prod)
|
||||
self.logger.record("social/delta_productivity", trend_pord)
|
||||
|
||||
@@ -241,7 +242,7 @@ econ=foundation.make_env_instance(**env_config)
|
||||
|
||||
market=econ.get_component("ContinuousDoubleAuction")
|
||||
action=market.get_n_actions("TradingAgent")
|
||||
baseEconWrapper=BaseEconWrapper(econ)
|
||||
baseEconWrapper=BaseEconWrapper(econ,True)
|
||||
baseEconWrapper.run()
|
||||
time.sleep(0.5)
|
||||
mobileRecieverEconWrapper=RecieverEconWrapper(base_econ=baseEconWrapper,agent_classname="BasicMobileAgent")
|
||||
@@ -261,7 +262,7 @@ stackenv_traid=vec_frame_stack.VecFrameStack(venv=montraidingenv,n_stack=num_fra
|
||||
econ_eval=foundation.make_env_instance(**eval_env_config)
|
||||
|
||||
|
||||
baseEconWrapper_eval=BaseEconWrapper(econ_eval)
|
||||
baseEconWrapper_eval=BaseEconWrapper(econ_eval,False)
|
||||
baseEconWrapper_eval.run()
|
||||
time.sleep(0.5)
|
||||
mobileRecieverEconWrapper_eval=RecieverEconWrapper(base_econ=baseEconWrapper_eval,agent_classname="BasicMobileAgent")
|
||||
@@ -292,8 +293,10 @@ 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=learning_rate_adj_basic,env=stackenv_basic, seed=445,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=learning_rate_adj_trade,env=stackenv_traid, seed=445,verbose=1,device="cuda",tensorboard_log="./log")
|
||||
model = MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.001, vf_coef=0.5 ,gamma=0.99, learning_rate=4e-4,env=stackenv_basic, seed=500,verbose=1,device="cuda",tensorboard_log="./log")
|
||||
model_trade=MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.001, vf_coef=0.5 ,gamma=0.99, learning_rate=4e-4,env=stackenv_traid, seed=500,verbose=1,device="cuda",tensorboard_log="./log")
|
||||
model=model.load("basic.ai",stackenv_basic)
|
||||
model_trade=model_trade.load("trade.ai",stackenv_traid)
|
||||
|
||||
n_agents=econ.n_agents
|
||||
|
||||
@@ -308,9 +311,9 @@ while True:
|
||||
#Train
|
||||
runname="run_{}_{}".format(run_number,"basic")
|
||||
|
||||
thread_model=Thread(target=train,args=(model,total_required_for_episode_basic*150,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*150,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()
|
||||
@@ -324,6 +327,7 @@ while True:
|
||||
|
||||
## Run Eval
|
||||
print("### EVAL ###")
|
||||
baseEconWrapper_eval.force_reset()
|
||||
obs_basic=stackenv_basic_eval.reset()
|
||||
obs_trade=stackenv_traid_eval.reset()
|
||||
done=False
|
||||
|
||||
+2
-1
@@ -1,3 +1,4 @@
|
||||
|
||||
pandas
|
||||
pymongo
|
||||
gym
|
||||
ray[rllib]
|
||||
@@ -25,6 +25,7 @@ from stable_baselines3.common.callbacks import BaseCallback
|
||||
import yaml
|
||||
import time
|
||||
from threading import Thread
|
||||
from db.export_to_mongo import MongoExporter
|
||||
|
||||
env_config = {
|
||||
# ===== SCENARIO CLASS =====
|
||||
@@ -41,29 +42,30 @@ env_config = {
|
||||
# (1) Building houses
|
||||
('Craft', {'skill_dist': "pareto", 'commodities': ["Gem"],'max_skill_amount_benefit':1.5}),
|
||||
# (2) Trading collectible resources
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10}),
|
||||
('ContinuousDoubleAuction', {'max_num_orders': 10,'max_bid_ask':100}),
|
||||
# (3) Movement and resource collection
|
||||
('SimpleGather', {}),
|
||||
('ExternalMarket',{'market_demand':{
|
||||
'Gem': 15
|
||||
'Gem': 90
|
||||
}}),
|
||||
],
|
||||
|
||||
# ===== SCENARIO CLASS ARGUMENTS =====
|
||||
# (optional) kwargs that are added by the Scenario class (i.e. not defined in BaseEnvironment)
|
||||
|
||||
'starting_agent_coin': 10,
|
||||
'starting_agent_coin': 50,
|
||||
'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)
|
||||
'agent_composition': {"BasicMobileAgent": 2,"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
|
||||
'isoelastic_eta':0.001,
|
||||
'allow_observation_scaling': True,
|
||||
'dense_log_frequency': 100,
|
||||
'world_dense_log_frequency':1,
|
||||
'energy_cost':0,
|
||||
'energy_cost':0.2,
|
||||
'energy_warmup_method': "auto",
|
||||
'energy_warmup_constant': 4000,
|
||||
|
||||
@@ -80,63 +82,8 @@ env_config = {
|
||||
'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
|
||||
|
||||
num_frames=1
|
||||
CONNECTION_STRING = "mongodb://root:econ@localhost:27017/"
|
||||
class TensorboardCallback(BaseCallback):
|
||||
"""
|
||||
Custom callback for plotting additional values in tensorboard.
|
||||
@@ -161,6 +108,23 @@ class TensorboardCallback(BaseCallback):
|
||||
|
||||
return True
|
||||
|
||||
min_at_target_basic=0.5
|
||||
min_lr_basic=5e-6
|
||||
start_lr_basic=9e-4
|
||||
|
||||
min_at_target_trade=0.5
|
||||
min_lr_trade=5e-6
|
||||
start_lr_trade=9e-4
|
||||
|
||||
def learning_rate_adj_basic(x) -> float:
|
||||
diff=start_lr_basic-min_lr_basic
|
||||
lr=min_lr_basic+x*diff
|
||||
return lr
|
||||
|
||||
def learning_rate_adj_trade(x) -> float:
|
||||
diff=start_lr_trade-min_lr_trade
|
||||
lr=min_lr_basic+x*diff
|
||||
return lr
|
||||
|
||||
def printMarket(market):
|
||||
for i in range(len(market)):
|
||||
@@ -219,16 +183,15 @@ def printReplay(econ,agentid):
|
||||
|
||||
#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=BaseEconWrapper(econ,False)
|
||||
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)
|
||||
sb3_traderConverter=SB3EconConverter(tradeRecieverEconWrapper,econ,"TradingAgent",False)
|
||||
sb3Converter=SB3EconConverter(mobileRecieverEconWrapper,econ,"BasicMobileAgent",False)
|
||||
# attach sb3 wrappers
|
||||
|
||||
monenv=VecMonitor(venv=sb3Converter,info_keywords=["social/productivity","trend/productivity"])
|
||||
@@ -238,28 +201,8 @@ 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()
|
||||
|
||||
obs=montraidingenv.reset()
|
||||
|
||||
# define training functions
|
||||
def train(model,timesteps, econ_call,process_bar,name,db,index):
|
||||
@@ -273,66 +216,47 @@ 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")
|
||||
model = MaskablePPO("MlpPolicy",n_steps=int(env_config['episode_length']*2),ent_coef=0.1, vf_coef=0.5 ,gamma=0.99, learning_rate=learning_rate_adj_basic,env=stackenv_basic, seed=445,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=learning_rate_adj_trade,env=stackenv_traid, seed=445,verbose=1,device="cuda",tensorboard_log="./log")
|
||||
# Setup complete
|
||||
|
||||
n_agents=econ.n_agents
|
||||
# Load models
|
||||
model=model.load("basic.ai")
|
||||
model_trade=model_trade.load("trade.ai")
|
||||
|
||||
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))
|
||||
exporter=MongoExporter(CONNECTION_STRING,"econ",econ)
|
||||
|
||||
while True:
|
||||
|
||||
|
||||
#Train
|
||||
runname="run_{}_{}".format(run_number,"basic")
|
||||
|
||||
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*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_basic=stackenv_basic_eval.reset()
|
||||
obs_trade=stackenv_traid_eval.reset()
|
||||
baseEconWrapper.force_reset()
|
||||
obs_basic=stackenv_basic.reset()
|
||||
obs_trade=stackenv_traid.reset()
|
||||
done=False
|
||||
for i in tqdm(range(eval_env_config['episode_length'])):
|
||||
for i in tqdm(range(env_config['episode_length'])):
|
||||
#create masks
|
||||
masks_basic=stackenv_basic_eval.action_masks()
|
||||
masks_trade=stackenv_traid_eval.action_masks()
|
||||
masks_basic=stackenv_basic.action_masks()
|
||||
masks_trade=stackenv_traid.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])
|
||||
sb3Converter.step_async(action_basic[0])
|
||||
sb3_traderConverter.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])
|
||||
obs_basic,rew_basic,done_e,info=stackenv_basic.step(action_basic[0])
|
||||
obs_trade,rew_trade,done_e,info=stackenv_traid.step(action_trade[0])
|
||||
done=done_e[0]
|
||||
|
||||
|
||||
|
||||
market=econ_eval.get_component("ContinuousDoubleAuction")
|
||||
craft=econ_eval.get_component("Craft")
|
||||
exporter.submit_full_market()
|
||||
market=econ.get_component("ContinuousDoubleAuction")
|
||||
craft=econ.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)
|
||||
#printReplay(econ,5)
|
||||
#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"]))
|
||||
|
||||
Binary file not shown.
@@ -17,7 +17,7 @@ class BaseEconWrapper():
|
||||
info=None
|
||||
n_data_retrieved=0
|
||||
|
||||
def __init__(self, econ: base_env.BaseEnvironment):
|
||||
def __init__(self, econ: base_env.BaseEnvironment,auto_reset):
|
||||
self.env=econ
|
||||
self.vote_lock=Lock()
|
||||
|
||||
@@ -26,6 +26,8 @@ class BaseEconWrapper():
|
||||
self.action_edit_lock=Lock()
|
||||
self.stop_edit_lock=Lock()
|
||||
self.env_data_lock=Lock()
|
||||
self.auto_reset=auto_reset
|
||||
|
||||
|
||||
def register_vote(self):
|
||||
"""Register reciever on base. Returns ID of Voter to pass on during blocking"""
|
||||
@@ -79,7 +81,7 @@ class BaseEconWrapper():
|
||||
|
||||
#check for reset
|
||||
self.vote_lock.acquire() # we might edit votes
|
||||
if self.n_voters==self.n_votes_reset:
|
||||
if self.n_voters<=self.n_votes_reset:
|
||||
## perform reset
|
||||
self.n_votes_reset=0
|
||||
self._reset()
|
||||
@@ -96,6 +98,7 @@ class BaseEconWrapper():
|
||||
# release actions
|
||||
# we are done
|
||||
|
||||
|
||||
def stop_env(self):
|
||||
"""Stops the wrapper"""
|
||||
self.stop_edit_lock.acquire()
|
||||
@@ -110,13 +113,17 @@ class BaseEconWrapper():
|
||||
self.n_votes_reset=0
|
||||
self.obs=self.env.reset() #Reset env
|
||||
self.rew=None
|
||||
self.done=None
|
||||
self.done={'__all__': False}
|
||||
self.info=None
|
||||
self.env_data_lock.release() #Release lock
|
||||
# Notify for reset
|
||||
self.reset_notification.set()
|
||||
for v in self.step_notifications:
|
||||
v.clear() # unlock stepping
|
||||
def force_reset(self):
|
||||
"""Force a reset with no votes"""
|
||||
self._reset()
|
||||
#self.n_votes_reset=-self.n_voters
|
||||
|
||||
def _step(self):
|
||||
"""Steping interaly"""
|
||||
|
||||
@@ -21,6 +21,7 @@ class SB3EconConverter(VecEnv, gym.Env):
|
||||
obs0=utils.package(obs[0],*self.packager)
|
||||
obs0["flat"]
|
||||
self.step_request_send=False
|
||||
self.reset_request_send=False
|
||||
self.auto_reset=auto_reset
|
||||
self.observation_space=gym.spaces.Box(low=0,high=10,shape=(len(obs0["flat"]),),dtype=np.float32)
|
||||
super().__init__(self.num_envs, self.observation_space, self.action_space)
|
||||
@@ -73,7 +74,7 @@ class SB3EconConverter(VecEnv, gym.Env):
|
||||
|
||||
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():
|
||||
|
||||
Reference in New Issue
Block a user