1
0
mirror of https://github.com/gsi-upm/soil synced 2025-09-14 20:22:22 +00:00
Files
soil/soil/simulation.py
J. Fernando Sánchez 9749f4ca14 Fix multithreading
Multithreading needs pickling to work.
Pickling/unpickling didn't work in some situations, like when the
environment_agents parameter was left blank.
This was due to two reasons:

1) agents and history didn't have a setstate method, and some of their
attributes cannot be pickled (generators, sqlite connection)
2) the environment was adding generators (agents) to its state.

This fixes the situation by restricting the keys that the environment exports
when it pickles, and by adding the set/getstate methods in agents.

The resulting pickles should contain enough information to inspect
them (history, state values, etc), but very limited.
2018-12-09 16:58:49 +01:00

290 lines
12 KiB
Python

import os
import time
import importlib
import sys
import yaml
import traceback
import networkx as nx
from networkx.readwrite import json_graph
from multiprocessing import Pool
from functools import partial
import pickle
from nxsim import NetworkSimulation
from . import utils, basestring, agents
from .environment import Environment
from .utils import logger
class Simulation(NetworkSimulation):
"""
Subclass of nsim.NetworkSimulation with three main differences:
1) agent type can be specified by name or by class.
2) instead of just one type, a network agents distribution can be used.
The distribution specifies the weight (or probability) of each
agent type in the topology. This is an example distribution: ::
[
{'agent_type': 'agent_type_1',
'weight': 0.2,
'state': {
'id': 0
}
},
{'agent_type': 'agent_type_2',
'weight': 0.8,
'state': {
'id': 1
}
}
]
In this example, 20% of the nodes will be marked as type
'agent_type_1'.
3) if no initial state is given, each node's state will be set
to `{'id': 0}`.
Parameters
---------
name : str, optional
name of the Simulation
topology : networkx.Graph instance, optional
network_params : dict
parameters used to create a topology with networkx, if no topology is given
network_agents : dict
definition of agents to populate the topology with
agent_type : NetworkAgent subclass, optional
Default type of NetworkAgent to use for nodes not specified in network_agents
states : list, optional
List of initial states corresponding to the nodes in the topology. Basic form is a list of integers
whose value indicates the state
dir_path : str, optional
Directory path where to save pickled objects
seed : str, optional
Seed to use for the random generator
num_trials : int, optional
Number of independent simulation runs
max_time : int, optional
Time how long the simulation should run
environment_params : dict, optional
Dictionary of globally-shared environmental parameters
environment_agents: dict, optional
Similar to network_agents. Distribution of Agents that control the environment
environment_class: soil.environment.Environment subclass, optional
Class for the environment. It defailts to soil.environment.Environment
load_module : str, module name, deprecated
If specified, soil will load the content of this module under 'soil.agents.custom'
"""
def __init__(self, name=None, topology=None, network_params=None,
network_agents=None, agent_type=None, states=None,
default_state=None, interval=1, dump=None, dry_run=False,
dir_path=None, num_trials=1, max_time=100,
load_module=None, seed=None,
environment_agents=None, environment_params=None,
environment_class=None, **kwargs):
if topology is None:
topology = utils.load_network(network_params,
dir_path=dir_path)
elif isinstance(topology, basestring) or isinstance(topology, dict):
topology = json_graph.node_link_graph(topology)
self.load_module = load_module
self.topology = nx.Graph(topology)
self.network_params = network_params
self.name = name or 'UnnamedSimulation'
self.num_trials = num_trials
self.max_time = max_time
self.default_state = default_state or {}
self.dir_path = dir_path or os.getcwd()
self.interval = interval
self.seed = str(seed) or str(time.time())
self.dump = dump
self.dry_run = dry_run
sys.path += [self.dir_path, os.getcwd()]
self.environment_params = environment_params or {}
self.environment_class = utils.deserialize(environment_class,
known_modules=['soil.environment', ]) or Environment
environment_agents = environment_agents or []
self.environment_agents = agents._convert_agent_types(environment_agents,
known_modules=[self.load_module])
distro = agents.calculate_distribution(network_agents,
agent_type)
self.network_agents = agents._convert_agent_types(distro,
known_modules=[self.load_module])
self.states = agents._validate_states(states,
self.topology)
def run_simulation(self, *args, **kwargs):
return self.run(*args, **kwargs)
def run(self, *args, **kwargs):
return list(self.run_simulation_gen(*args, **kwargs))
def run_simulation_gen(self, *args, parallel=False, dry_run=False,
**kwargs):
p = Pool()
with utils.timer('simulation {}'.format(self.name)):
if parallel:
func = partial(self.run_trial_exceptions, dry_run=dry_run or self.dry_run,
return_env=True,
**kwargs)
for i in p.imap_unordered(func, range(self.num_trials)):
if isinstance(i, Exception):
logger.error('Trial failed:\n\t{}'.format(i.message))
continue
yield i
else:
for i in range(self.num_trials):
yield self.run_trial(i, dry_run = dry_run or self.dry_run, **kwargs)
if not (dry_run or self.dry_run):
logger.info('Dumping results to {}'.format(self.dir_path))
self.dump_pickle(self.dir_path)
self.dump_yaml(self.dir_path)
else:
logger.info('NOT dumping results')
def get_env(self, trial_id = 0, **kwargs):
opts=self.environment_params.copy()
env_name='{}_trial_{}'.format(self.name, trial_id)
opts.update({
'name': env_name,
'topology': self.topology.copy(),
'seed': self.seed+env_name,
'initial_time': 0,
'dry_run': self.dry_run,
'interval': self.interval,
'network_agents': self.network_agents,
'states': self.states,
'default_state': self.default_state,
'environment_agents': self.environment_agents,
'dir_path': self.dir_path,
})
opts.update(kwargs)
env=self.environment_class(**opts)
return env
def run_trial(self, trial_id = 0, until = None, return_env = True, **opts):
"""Run a single trial of the simulation
Parameters
----------
trial_id : int
"""
# Set-up trial environment and graph
until=until or self.max_time
env=self.get_env(trial_id = trial_id, **opts)
# Set up agents on nodes
with utils.timer('Simulation {} trial {}'.format(self.name, trial_id)):
env.run(until)
if self.dump and not self.dry_run:
with utils.timer('Dumping simulation {} trial {}'.format(self.name, trial_id)):
env.dump(formats = self.dump)
if return_env:
return env
def run_trial_exceptions(self, *args, **kwargs):
'''
A wrapper for run_trial that catches exceptions and returns them.
It is meant for async simulations
'''
try:
return self.run_trial(*args, **kwargs)
except Exception as ex:
c = ex.__cause__
c.message = ''.join(traceback.format_exception(type(c), c, c.__traceback__)[:])
return c
def to_dict(self):
return self.__getstate__()
def to_yaml(self):
return yaml.dump(self.to_dict())
def dump_yaml(self, dir_path = None, file_name = None):
dir_path=dir_path or self.dir_path
if not os.path.exists(dir_path):
os.makedirs(dir_path)
if not file_name:
file_name=os.path.join(dir_path,
'{}.dumped.yml'.format(self.name))
with open(file_name, 'w') as f:
f.write(self.to_yaml())
def dump_pickle(self, dir_path = None, pickle_name = None):
dir_path=dir_path or self.dir_path
if not os.path.exists(dir_path):
os.makedirs(dir_path)
if not pickle_name:
pickle_name=os.path.join(dir_path,
'{}.simulation.pickle'.format(self.name))
with open(pickle_name, 'wb') as f:
pickle.dump(self, f)
def __getstate__(self):
state={}
for k, v in self.__dict__.items():
if k[0] != '_':
state[k]=v
state['topology']=json_graph.node_link_data(self.topology)
state['network_agents']=agents.serialize_distribution(self.network_agents,
known_modules = [])
state['environment_agents']=agents.serialize_distribution(self.environment_agents,
known_modules = [])
state['environment_class']=utils.serialize(self.environment_class,
known_modules=['soil.environment'])[1] # func, name
if state['load_module'] is None:
del state['load_module']
return state
def __setstate__(self, state):
self.__dict__ = state
self.load_module = getattr(self, 'load_module', None)
if self.dir_path not in sys.path:
sys.path += [self.dir_path, os.getcwd()]
self.topology = json_graph.node_link_graph(state['topology'])
self.network_agents = agents.calculate_distribution(agents._convert_agent_types(self.network_agents))
self.environment_agents = agents._convert_agent_types(self.environment_agents,
known_modules=[self.load_module])
self.environment_class = utils.deserialize(self.environment_class,
known_modules=[self.load_module, 'soil.environment', ]) # func, name
return state
def from_config(config):
config = list(utils.load_config(config))
if len(config) > 1:
raise AttributeError('Provide only one configuration')
config = config[0][0]
sim = Simulation(**config)
return sim
def run_from_config(*configs, results_dir='soil_output', dump=None, timestamp=False, **kwargs):
for config_def in configs:
# logger.info("Found {} config(s)".format(len(ls)))
for config, _ in utils.load_config(config_def):
name = config.get('name', 'unnamed')
logger.info("Using config(s): {name}".format(name=name))
if timestamp:
sim_folder = '{}_{}'.format(name,
time.strftime("%Y-%m-%d_%H:%M:%S"))
else:
sim_folder = name
dir_path = os.path.join(results_dir, sim_folder)
if dump is not None:
config['dump'] = dump
sim = Simulation(dir_path=dir_path, **config)
logger.info('Dumping results to {} : {}'.format(sim.dir_path, sim.dump))
sim.run_simulation(**kwargs)