def write_metric_logs(metric):
"""Write metric logs"""
keys = []
log, _ = format_custom_logs(keys=keys,
raw_log=flatten_dict(metric),
log_type="metric")
write_log(log)
def write_metric_logs(self, metrics):
"""Write Metric"""
metrics["experiment_id"] = self._experiment_id
fs_log.write_metric_logs(metrics)
flattened_metrics = flatten_dict(metrics, sep="_")
if self.metrics_to_omit:
metric_dict = {
key: flattened_metrics[key]
for key in flattened_metrics if key not in self.metrics_to_omit
}
else:
metric_dict = flattened_metrics
prefix = metrics.get("mode", None)
num_timesteps = metric_dict.pop("minibatch")
# print(metric_dict)
self._log_metrics(dic=metric_dict, prefix=prefix, step=num_timesteps)
if self.should_use_tb:
timestep_key = "num_timesteps"
for key in set(list(metrics.keys())) - set([timestep_key]):
self.tensorboard_writer.add_scalar(
tag=key,
scalar_value=metrics[key],
global_step=metrics[timestep_key],
)
def __init__(self, config):
self._experiment_id = 0
self.metrics_to_omit = ["mode"]
flattened_config = flatten_dict(config.to_serializable_dict(), sep="_")
self.should_use_tb = False
fs_log.set_logger(config=config)
def write_compute_logs(self, **kwargs):
"""Write Compute Logs"""
kwargs["experiment_id"] = self._experiment_id
fs_log.write_metric_logs(**kwargs)
metric_dict = flatten_dict(kwargs, sep="_")
num_timesteps = metric_dict.pop("num_timesteps")
self._log_metrics(dic=metric_dict,
step=num_timesteps,
prefix="compute")
def write_config_log(self, config):
"""Write config"""
fs_log.write_config_log(config)
flatten_config = flatten_dict(config, sep="_")
flatten_config["experiment_id"] = self._experiment_id