def __init__(self):
"""
Default constructor.
"""
super(BaseAgent, self).__init__()
self._log = logger.get_logger(self.__class__.__name__)
self._load_configuration()
def __init__(self):
"""
Default constructor.
"""
self._config_handler = KatherineApplication.get_application_config()
if not self._config_handler:
raise ValueError("No config_handler specified")
self._log = logger.get_logger(self.__class__.__name__)
def __init__(self, config_uri: str):
"""
Default constructor.
:param config_uri:
RFC-3986 Uniform Resource Identifier (URI)
"""
self._log = logger.get_logger(self.__class__.__name__)
self.yaml = ConfigProcessor()
self.settings = {}
self._load_configuration(config_uri)
def __init__(self, name: Optional[str] = None):
"""Creates an instance of `Network`.
Args:
name: A string representing the name of the network.
"""
self._log = logger.get_logger(self.__class__.__name__)
self._config_handler = KatherineApplication.get_application_config()
self._serializer = KatherineApplication.get_application_factory(
).build_model_serializer()
self._name = name or None
self._load_configuration()
def __init__(self):
"""
Default constructor.
"""
self._log = logger.get_logger(self.__class__.__name__)
self._config_handler = KatherineApplication.get_application_config()
if not self._config_handler:
raise ValueError("No config specified.")
self._game = KatherineApplication.get_application_factory().build_game()
self._agent = KatherineApplication.get_application_factory().build_agent()
self._metrics = KatherineApplication.get_application_factory().build_metrics_tracer()
self._load_configuration()
def __init__(self):
"""
Default constructor.
"""
self._log = logger.get_logger(self.__class__.__name__)
config_handler = KatherineApplication.get_application_config()
self.enabled_metrics = config_handler.get_config_property(
TensorBoardConfigurationProperty.ENABLED_NETWORK_METRICS,
TensorBoardConfigurationProperty.ENABLED_NETWORK_METRICS.prop_type)
self.is_profiler_enabled = config_handler.get_config_property(
TensorBoardConfigurationProperty.GPU_PROFILER_ENABLED,
TensorBoardConfigurationProperty.GPU_PROFILER_ENABLED.prop_type)
self.work_directory = fileio.build_metrics_work_directory(WORKING_DIRECTORY_PREFIX)
self.initialized = False
# | \ / _` | __| '_ \ / _ \ '__| | '_ \ / _ \ #
# | |\ \ (_| | |_| | | | __/ | | | | | | __/ #
# \_| \_/\__,_|\__|_| |_|\___|_| |_|_| |_|\___| #
# #
# General Video Game AI #
# Copyright (C) 2020-2021 d33are #
##################################################
from kat_api import IPropertyDescriptor, ITensorDescriptor, IObservation
from kat_typing import Tensor
from kat_framework.core.descriptors import TensorDescriptor
from kat_framework.util import logger
from typing import Collection
import numpy as np
log = logger.get_logger(__name__)
def check_same_tensor_structure(input_tensor: Tensor,
property_desc: ITensorDescriptor,
reduce_batch_dim: bool = False) -> bool:
"""
Checks that the specified tensor is compatible or not with the provided descriptor.
:param input_tensor:
tensor to check
:param property_desc:
descriptor checked by
:param reduce_batch_dim:
batch dimension is needed or not
:return:
# _ __ _ _ _ #
# | | / / | | | | (_) #
# | |/ / __ _| |_| |__ ___ _ __ _ _ __ ___ #
# | \ / _` | __| '_ \ / _ \ '__| | '_ \ / _ \ #
# | |\ \ (_| | |_| | | | __/ | | | | | | __/ #
# \_| \_/\__,_|\__|_| |_|\___|_| |_|_| |_|\___| #
# #
# General Video Game AI #
# Copyright (C) 2020-2021 d33are #
##################################################
from kat_api import IReadOnlyMemory, IReplayMemory
from kat_typing import IterableDataset
from kat_framework.util import logger
log = logger.get_logger(__name__ + ".MemoryAccessor")
class MemoryAccessor(IReadOnlyMemory):
"""
Read only replay memory wrapper.
"""
# protected members
_replay_memory: IReplayMemory = None
# public members
def __init__(self):
"""
def __init__(self):
"""
Default constructor.
"""
super(UniformMemory, self).__init__()
self._log = logger.get_logger(self.__class__.__name__)
from kat_framework.util import reflection, logger
LOGO_STRING = \
"\n ##################################################\n \
# _ __ _ _ _ #\n \
# | | / / | | | | (_) #\n \
# | |/ / __ _| |_| |__ ___ _ __ _ _ __ ___ #\n \
# | \\ / _` | __| '_ \\ / _ \\ '__| | '_ \\ / _ \\ #\n \
# | |\\ \\ (_| | |_| | | | __/ | | | | | | __/ #\n \
# \\_| \\_/\\__,_|\\__|_| |_|\\___|_| |_|_| |_|\\___| #\n \
# #\n \
# General Video Game AI #\n \
# Copyright (C) 2020-2021 d33are #\n \
##################################################\n"
log = logger.get_logger(__name__ + ".KatherineApplication")
class KatherineApplication:
"""
Katherine application entrypoint.
"""
application_factory: IFactory
config_handler: IConfigurationHandler
@staticmethod
def init(factory_class: str, config_handler_class: str,
config_uri: str) -> None:
"""
Runtime initialization.
class TensorboardTracer(IMetricTracer):
"""
Tensorflow based metrics collector implementation.
# see : IMetricTracer
"""
_log = logger.get_logger(__name__ + ".TensorboardTracer")
config_handler: IConfigurationHandler = None
enabled_metrics: Collection[KatMetrics] = None
is_profiler_enabled: bool = False
work_directory: str = None
strategy: DistributionStrategy = None
metrics: dict = None
summary_writer: tf.summary.SummaryWriter = None
initialized: bool = False
def __init__(self):
"""
Default constructor.
"""
config_handler = KatherineApplication.get_application_config()
self.enabled_metrics = config_handler.get_config_property(
TensorBoardConfigurationProperty.ENABLED_NETWORK_METRICS,
TensorBoardConfigurationProperty.ENABLED_NETWORK_METRICS.prop_type)
self.is_profiler_enabled = config_handler.get_config_property(
TensorBoardConfigurationProperty.GPU_PROFILER_ENABLED,
TensorBoardConfigurationProperty.GPU_PROFILER_ENABLED.prop_type)
self.work_directory = fileio.build_metrics_work_directory(WORKING_DIRECTORY_PREFIX)
@overrides
def is_initialized(self) -> bool:
"""
Initialization flag.
# see : IMetricTracer.is_initialized()
"""
return self.initialized
@overrides
def init(self, distribution_strategy: Optional[DistributionStrategy] = None):
"""
Object initialization.
# see : IMetricTracer.init(distribution_strategy)
"""
if distribution_strategy is None:
self.strategy = tf.distribute.get_strategy()
else:
self.strategy = distribution_strategy
with self.strategy.scope():
self.metrics = self._build_metrics()
self.summary_writer = tf.summary.create_file_writer(self.work_directory)
self.initialized = True
@overrides
def start_profiler(self):
"""
# see : IMetricTracer.start_profiler()
"""
if self.is_profiler_enabled:
tf.profiler.experimental.start(self.work_directory)
@overrides
def stop_profiler(self):
"""
# see : IMetricTracer.stop_profiler()
"""
if self.is_profiler_enabled:
tf.profiler.experimental.stop()
@overrides
def flush_metrics(self, epoch_number: int) -> None:
"""
# see : IMetricTracer.flush_metrics(epoch_number)
"""
with self.summary_writer.as_default():
for key, value in self.metrics.items():
if MetricType.SCALAR == key.metric_type:
if key.metric_clazz is None:
tf.summary.scalar(key.label, data=value, step=epoch_number)
else:
tf.summary.scalar(key.label, value.result(), step=epoch_number)
elif MetricType.IMAGE == key.metric_type:
tf.summary.image(key.label, value, step=epoch_number)
elif MetricType.HYPER_PARAMETER == key.metric_type:
raise NotImplemented
elif MetricType.FAIRNESS_INDICATOR == key.metric_type:
raise NotImplemented
elif MetricType.EMBEDDING == key.metric_type:
raise NotImplemented
@overrides
def update_metric(self, metadata: TraceableMetric, data: MetricData) -> None:
"""
# see : IMetricTracer.update_metric(metadata, data)
"""
if metadata is None:
raise ValueError("No metadata specified")
if data is None:
raise ValueError("No data specified")
metric = self.metrics.get(metadata)
if metric is not None and metadata.metric_clazz is not None:
metric(data)
elif metric is not None:
self.metrics[metadata] = data
else:
if self._log.isEnabledFor(DEBUG):
self._log.debug("For metric {} the tracing is not enabled from configuration.".format(metadata))
# protected member functions
def _build_metrics(self):
"""
Builds the metrics dictionary based on the provided metadata.
"""
return dict(map(lambda m: (m, metrics.build_metric_from_metadata(m, tf.keras.metrics.Metric)),
self.enabled_metrics))