如何在分布式环境中使用Estimator API在Tensorboard中显示运行时统计信息

import tensorflow as tf
from tensorflow.estimator import Estimator
from tensorflow.estimator.inputs import numpy_input_fn
from tensorflow.python.training import device_setter

class MyEstimator(Estimator):
    def __init__(self, model_dir=None, config=None, params=None):
        super(MyEstimator, self).__init__(model_dir=model_dir, config=config, params=params)

    def model_fn(self, features, labels, mode, params):
        # 定义模型的结构和计算图
        ...

        # 定义损失函数和优化器
        ...

        # 定义评估指标
        ...

        # 返回EstimatorSpec对象
        return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op, eval_metric_ops=eval_metric_ops)

config = tf.estimator.RunConfig(
    model_dir=model_dir,
    save_summary_steps=100,
    save_checkpoints_steps=1000,
    session_config=tf.ConfigProto(allow_soft_placement=True),
    train_distribute=tf.contrib.distribute.ParameterServerStrategy(),
    eval_distribute=tf.contrib.distribute.MirroredStrategy()
)

estimator = MyEstimator(model_dir=model_dir, config=config, params=params)

train_spec = tf.estimator.TrainSpec(input_fn=train_input_fn, max_steps=num_train_steps)
eval_spec = tf.estimator.EvalSpec(input_fn=eval_input_fn, steps=num_eval_steps)

tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)

summary_hook = tf.estimator.SummarySaverHook(
    save_steps=100,
    output_dir=model_dir,
    summary_op=tf.summary.merge_all()
)

logging_hook = tf.train.LoggingTensorHook(
    tensors={"loss": loss, "accuracy": accuracy},
    every_n_iter=100
)

estimator.train(
    input_fn=train_input_fn,
    steps=num_train_steps,
    hooks=[summary_hook, logging_hook]
)

tensorboard --logdir=model_dir