mirror of
https://github.com/illiumst/marl-factory-grid.git
synced 2025-07-06 01:21:36 +02:00
Merge branch 'main' into unit_testing
This commit is contained in:
@ -1,4 +1,5 @@
|
||||
import ast
|
||||
from collections import defaultdict
|
||||
|
||||
from os import PathLike
|
||||
from pathlib import Path
|
||||
@ -24,13 +25,21 @@ class FactoryConfigParser(object):
|
||||
self.config_path = Path(config_path)
|
||||
self.custom_modules_path = Path(custom_modules_path) if custom_modules_path is not None else custom_modules_path
|
||||
self.config = yaml.safe_load(self.config_path.open())
|
||||
self._n_abbr_dict = None
|
||||
|
||||
def __getattr__(self, item):
|
||||
return self['General'][item]
|
||||
|
||||
def _get_sub_list(self, primary_key: str, sub_key: str):
|
||||
return [{key: [s for k, v in val.items() if k == sub_key for s in v] for key, val in x.items()
|
||||
} for x in self.config[primary_key]]
|
||||
} for x in self.config.get(primary_key, [])]
|
||||
|
||||
def _n_abbr(self, n):
|
||||
assert isinstance(n, int)
|
||||
if self._n_abbr_dict is None:
|
||||
self._n_abbr_dict = defaultdict(lambda: 'th', {1: 'st', 2: 'nd', 3: 'rd'})
|
||||
return self._n_abbr_dict[n]
|
||||
|
||||
|
||||
@property
|
||||
def agent_actions(self):
|
||||
@ -145,11 +154,18 @@ class FactoryConfigParser(object):
|
||||
observations.extend(x for x in self.agents[name]['Observations'] if x != c.DEFAULTS)
|
||||
positions = [ast.literal_eval(x) for x in self.agents[name].get('Positions', [])]
|
||||
other_kwargs = {k: v for k, v in self.agents[name].items() if k not in
|
||||
['Actions', 'Observations', 'Positions']}
|
||||
['Actions', 'Observations', 'Positions', 'Clones']}
|
||||
parsed_agents_conf[name] = dict(
|
||||
actions=parsed_actions, observations=observations, positions=positions, other=other_kwargs
|
||||
)
|
||||
|
||||
clones = self.agents[name].get('Clones', 0)
|
||||
if clones:
|
||||
if isinstance(clones, int):
|
||||
clones = [f'{name}_the_{n}{self._n_abbr(n)}' for n in range(clones)]
|
||||
for clone in clones:
|
||||
parsed_agents_conf[clone] = parsed_agents_conf[name].copy()
|
||||
|
||||
return parsed_agents_conf
|
||||
|
||||
def load_env_rules(self) -> List[Rule]:
|
||||
|
@ -58,3 +58,6 @@ class EnvMonitor(Wrapper):
|
||||
pickle.dump(self._monitor_df.reset_index(), f, protocol=pickle.HIGHEST_PROTOCOL)
|
||||
if auto_plotting_keys:
|
||||
plot_single_run(filepath, column_keys=auto_plotting_keys)
|
||||
|
||||
def report_possible_colum_keys(self):
|
||||
print(self._monitor_df.columns)
|
@ -24,11 +24,7 @@ class OBSBuilder(object):
|
||||
return 0
|
||||
|
||||
def __init__(self, level_shape: np.size, state: Gamestate, pomdp_r: int):
|
||||
self._curr_env_step = None
|
||||
self.all_obs = dict()
|
||||
self.light_blockers = defaultdict(lambda: False)
|
||||
self.positional = defaultdict(lambda: False)
|
||||
self.non_positional = defaultdict(lambda: False)
|
||||
self.ray_caster = dict()
|
||||
|
||||
self.level_shape = level_shape
|
||||
@ -37,13 +33,15 @@ class OBSBuilder(object):
|
||||
self.size = np.prod(self.obs_shape)
|
||||
|
||||
self.obs_layers = dict()
|
||||
|
||||
self.reset_struc_obs_block(state)
|
||||
self.curr_lightmaps = dict()
|
||||
|
||||
self._floortiles = defaultdict(list, {pos: [Floor(*pos)] for pos in state.entities.floorlist})
|
||||
|
||||
def reset_struc_obs_block(self, state):
|
||||
self._curr_env_step = state.curr_step
|
||||
self.reset(state)
|
||||
|
||||
def reset(self, state):
|
||||
# Reset temporary information
|
||||
self.curr_lightmaps = dict()
|
||||
# Construct an empty obs (array) for possible placeholders
|
||||
self.all_obs[c.PLACEHOLDER] = np.full(self.obs_shape, 0, dtype=float)
|
||||
# Fill the all_obs-dict with all available entities
|
||||
@ -52,7 +50,8 @@ class OBSBuilder(object):
|
||||
|
||||
def observation_space(self, state):
|
||||
from gymnasium.spaces import Tuple, Box
|
||||
obsn = self.refresh_and_build_for_all(state)
|
||||
self.reset(state)
|
||||
obsn = self.build_for_all(state)
|
||||
if len(state[c.AGENT]) == 1:
|
||||
space = Box(low=0, high=1, shape=next(x for x in obsn.values()).shape, dtype=np.float32)
|
||||
else:
|
||||
@ -60,14 +59,13 @@ class OBSBuilder(object):
|
||||
return space
|
||||
|
||||
def named_observation_space(self, state):
|
||||
return self.refresh_and_build_for_all(state)
|
||||
self.reset(state)
|
||||
return self.build_for_all(state)
|
||||
|
||||
def refresh_and_build_for_all(self, state) -> (dict, dict):
|
||||
self.reset_struc_obs_block(state)
|
||||
def build_for_all(self, state) -> (dict, dict):
|
||||
return {agent.name: self.build_for_agent(agent, state)[0] for agent in state[c.AGENT]}
|
||||
|
||||
def refresh_and_build_named_for_all(self, state) -> Dict[str, Dict[str, np.ndarray]]:
|
||||
self.reset_struc_obs_block(state)
|
||||
def build_named_for_all(self, state) -> Dict[str, Dict[str, np.ndarray]]:
|
||||
named_obs_dict = {}
|
||||
for agent in state[c.AGENT]:
|
||||
obs, names = self.build_for_agent(agent, state)
|
||||
@ -85,9 +83,6 @@ class OBSBuilder(object):
|
||||
pass
|
||||
|
||||
def build_for_agent(self, agent, state) -> (List[str], np.ndarray):
|
||||
assert self._curr_env_step == state.curr_step, (
|
||||
"The observation objekt has not been reset this state! Call 'reset_struc_obs_block(state)'"
|
||||
)
|
||||
try:
|
||||
agent_want_obs = self.obs_layers[agent.name]
|
||||
except KeyError:
|
||||
@ -166,7 +161,8 @@ class OBSBuilder(object):
|
||||
raise ValueError(f'Max(obs.size) for {e.name}: {obs[idx].size}, but was: {len(v)}.')
|
||||
if self.pomdp_r:
|
||||
try:
|
||||
light_map = np.zeros(self.obs_shape)
|
||||
light_map = self.curr_lightmaps.get(agent.name, np.zeros(self.obs_shape))
|
||||
light_map[:] = 0.0
|
||||
visible_floor = self.ray_caster[agent.name].visible_entities(self._floortiles, reset_cache=False)
|
||||
|
||||
for f in set(visible_floor):
|
||||
|
@ -49,7 +49,7 @@ def prepare_plt(df, hue, style, hue_order):
|
||||
plt.close('all')
|
||||
sns.set(rc={'text.usetex': False}, style='whitegrid')
|
||||
lineplot = sns.lineplot(data=df, x='Episode', y='Score', hue=hue, style=style,
|
||||
ci=95, palette=PALETTE, hue_order=hue_order, )
|
||||
errorbar=('ci', 95), palette=PALETTE, hue_order=hue_order, )
|
||||
plt.legend(bbox_to_anchor=(1.02, 1), loc='upper left', borderaxespad=0)
|
||||
plt.tight_layout()
|
||||
# lineplot.set_title(f'{sorted(list(df["Measurement"].unique()))}')
|
||||
|
@ -8,7 +8,7 @@ import numpy as np
|
||||
from marl_factory_grid.algorithms.static.utils import points_to_graph
|
||||
from marl_factory_grid.environment import constants as c
|
||||
from marl_factory_grid.environment.entity.entity import Entity
|
||||
from marl_factory_grid.environment.rules import Rule
|
||||
from marl_factory_grid.environment.rules import Rule, SpawnAgents
|
||||
from marl_factory_grid.utils.results import Result, DoneResult
|
||||
from marl_factory_grid.environment.tests import Test
|
||||
from marl_factory_grid.utils.results import Result
|
||||
@ -32,18 +32,19 @@ class StepRules:
|
||||
self.rules.append(item)
|
||||
return True
|
||||
|
||||
def do_all_reset(self, state):
|
||||
for rule in self.rules:
|
||||
if rule_reset_printline := rule.on_reset(state):
|
||||
state.print(rule_reset_printline)
|
||||
return c.VALID
|
||||
|
||||
def do_all_init(self, state, lvl_map):
|
||||
for rule in self.rules:
|
||||
if rule_init_printline := rule.on_init(state, lvl_map):
|
||||
state.print(rule_init_printline)
|
||||
return c.VALID
|
||||
|
||||
def do_all_reset(self, state):
|
||||
SpawnAgents().on_reset(state)
|
||||
for rule in self.rules:
|
||||
if rule_reset_printline := rule.on_reset(state):
|
||||
state.print(rule_reset_printline)
|
||||
return c.VALID
|
||||
|
||||
def tick_step_all(self, state):
|
||||
results = list()
|
||||
for rule in self.rules:
|
||||
@ -91,6 +92,10 @@ class Gamestate(object):
|
||||
self._floortile_graph = None
|
||||
self.tests = StepTests(*tests)
|
||||
|
||||
def reset(self):
|
||||
self.curr_step = 0
|
||||
self.curr_actions = None
|
||||
|
||||
def __getitem__(self, item):
|
||||
return self.entities[item]
|
||||
|
||||
@ -201,7 +206,7 @@ class Gamestate(object):
|
||||
results.extend(on_check_done_result)
|
||||
return results
|
||||
|
||||
def get_all_pos_with_collisions(self) -> List[Tuple[(int, int)]]:
|
||||
def get_collision_positions(self) -> List[Tuple[(int, int)]]:
|
||||
"""
|
||||
Returns a list positions [(x, y), ... ] on which collisions occur. This does not include agents,
|
||||
that were unable to move because their target direction was blocked, also a form of collision.
|
||||
|
Reference in New Issue
Block a user