steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-05-23 15:26:43 +02:00
Code Issues Packages Projects Releases Wiki Activity

Monitor and Agent State Debug

Browse Source
This commit is contained in:
steffen-illium 2021-05-14 13:39:24 +02:00
parent 18be35dc19
commit 35ef708b20
2 changed files with 32 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
environments/factory/base_factory.py
View File

@ -13,9 +13,9 @@ class AgentState:
self.i = i self.i = i
self.action = action self.action = action
self.pos = None
self.collision_vector = None self.collision_vector = None
self.action_valid = None self.action_valid = None
self.pos = None
@property @property
def collisions(self): def collisions(self):
@ -34,6 +34,7 @@ class FactoryMonitor:
def __init__(self, env): def __init__(self, env):
self._env = env self._env = env
self._monitor = defaultdict(lambda: defaultdict(lambda: 0)) self._monitor = defaultdict(lambda: defaultdict(lambda: 0))
self._last_vals = defaultdict(lambda: 0)
def __iter__(self): def __iter__(self):
for key, value in self._monitor.items(): for key, value in self._monitor.items():
@ -42,19 +43,23 @@ class FactoryMonitor:
def add(self, key, value, step=None): def add(self, key, value, step=None):
assert step is None or step >= 1 # Is this good practice? assert step is None or step >= 1 # Is this good practice?
step = step or self._env.steps step = step or self._env.steps
self._monitor[key][step] = list(self._monitor[key].values())[-1] + value self._last_vals[key] = self._last_vals[key] + value
return self._monitor[key][step] self._monitor[key][step] = self._last_vals[key]
return self._last_vals[key]
def set(self, key, value, step=None): def set(self, key, value, step=None):
assert step is None or step >= 1 # Is this good practice? assert step is None or step >= 1 # Is this good practice?
step = step or self._env.steps step = step or self._env.steps
self._monitor[key][step] = value self._last_vals[key] = value
return self._monitor[key][step] self._monitor[key][step] = self._last_vals[key]
return self._last_vals[key]
def reduce(self, key, value, step=None): def remove(self, key, value, step=None):
assert step is None or step >= 1 # Is this good practice? assert step is None or step >= 1 # Is this good practice?
step = step or self._env.steps step = step or self._env.steps
self._monitor[key][step] = list(self._monitor[key].values())[-1] - value self._last_vals[key] = self._last_vals[key] - value
self._monitor[key][step] = self._last_vals[key]
return self._last_vals[key]
def to_dict(self): def to_dict(self):
return dict(self) return dict(self)
@ -83,13 +88,13 @@ class BaseFactory:
h.parse_level(Path(__file__).parent / h.LEVELS_DIR / f'{level}.txt') h.parse_level(Path(__file__).parent / h.LEVELS_DIR / f'{level}.txt')
) )
self.slice_strings = {0: 'level', **{i: f'agent#{i}' for i in range(1, self.n_agents+1)}} self.slice_strings = {0: 'level', **{i: f'agent#{i}' for i in range(1, self.n_agents+1)}}
self.monitor = FactoryMonitor(self)
self.reset() self.reset()
def reset(self): def reset(self):
self.done = False self.done = False
self.steps = 0 self.steps = 0
self.cumulative_reward = 0 self.cumulative_reward = 0
self.monitor = FactoryMonitor(self)
# Agent placement ... # Agent placement ...
agents = np.zeros((self.n_agents, *self.level.shape), dtype=np.int8) agents = np.zeros((self.n_agents, *self.level.shape), dtype=np.int8)
floor_tiles = np.argwhere(self.level == h.IS_FREE_CELL) floor_tiles = np.argwhere(self.level == h.IS_FREE_CELL)

30
environments/factory/simple_factory.py
View File

@ -1,5 +1,5 @@
import numpy as np import numpy as np
from environments.factory.base_factory import BaseFactory from environments.factory.base_factory import BaseFactory, FactoryMonitor
class SimpleFactory(BaseFactory): class SimpleFactory(BaseFactory):
@ -14,28 +14,36 @@ class SimpleFactory(BaseFactory):
self.state[-1, x, y] = 1 self.state[-1, x, y] = 1
def reset(self): def reset(self):
super().reset() state, r, done, _ = super().reset()
dirt_slice = np.zeros((1, *self.state.shape[1:])) dirt_slice = np.zeros((1, *self.state.shape[1:]))
self.state = np.concatenate((self.state, dirt_slice)) # dirt is now the last slice self.state = np.concatenate((self.state, dirt_slice)) # dirt is now the last slice
self.spawn_dirt() self.spawn_dirt()
# Always: This should return state, r, done, info
return self.state, r, done, _
def calculate_reward(self, agent_states): def calculate_reward(self, agent_states):
for agent_state in agent_states: for agent_state in agent_states:
collisions = agent_state.collisions collisions = agent_state.collisions
entities = [self.slice_strings[entity] for entity in collisions] entities = [self.slice_strings[entity] for entity in collisions]
for entity in entities: if entities:
self.monitor.add(f'{entity}_collisions', 1) for entity in entities:
print(f't = {self.steps}\tAgent {agent_state.i} has collisions with ' self.monitor.add(f'agent_{agent_state.i}_collision_{entity}', 1)
f'{entities}') print(f't = {self.steps}\tAgent {agent_state.i} has collisions with '
f'{entities}')
return 0, {} return 0, {}
if __name__ == '__main__': if __name__ == '__main__':
import random import random
factory = SimpleFactory(n_agents=1, max_dirt=8) factory = SimpleFactory(n_agents=1, max_dirt=8)
random_actions = [random.randint(0, 7) for _ in range(200)] monitor_list = list()
for action in random_actions: for epoch in range(100):
state, r, done, _ = factory.step(action) random_actions = [random.randint(0, 7) for _ in range(200)]
print(f'Factory run done, reward is:\n {r}') state, r, done, _ = factory.reset()
print(f'There have been the following collisions: \n {dict(factory.monitor)}') for action in random_actions:
state, r, done, info = factory.step(action)
monitor_list.append(factory.monitor)
print(f'Factory run done, reward is:\n {r}')
print(f'There have been the following collisions: \n {dict(factory.monitor)}')