def setup_logger(out_dir="results", exp_name="test", output_formats=None):
    timestamp = utils.make_timestamp()
    exp_name = exp_name.replace("/", "_")  # environment names can contain /'s
    filename = "{}-{}".format(timestamp, exp_name)[0:255]  # Linux has filename limit of 255
    out_dir = osp.join(out_dir, filename)
    os.makedirs(out_dir, exist_ok=True)

    logger.configure(folder=osp.join(out_dir, "rl"), format_strs=["tensorboard", "stdout"])
    logger_instance = logger.Logger.CURRENT

    if output_formats is not None:
        logger_instance.output_formats += output_formats

    for fmt in logger_instance.output_formats:
        if isinstance(fmt, logger.TensorBoardOutputFormat):
            writer = fmt.writer
            layout = tb_layout()
            event = event_pb2.Event(summary=layout)
            writer.WriteEvent(event)
            writer.Flush()

    return out_dir, logger_instance 

def load_tensor_from_event_file(event_file_path):
  """Load a tensor from an event file.

  Assumes that the event file contains a `Event` protobuf and the `Event`
  protobuf contains a `Tensor` value.

  Args:
    event_file_path: (`str`) path to the event file.

  Returns:
    The tensor value loaded from the event file, as a `numpy.ndarray`. For
    uninitialized Tensors, returns `None`. For Tensors of data types that
    cannot be converted to `numpy.ndarray` (e.g., `tf.resource`), return
    `None`.
  """

  event = event_pb2.Event()
  with gfile.Open(event_file_path, "rb") as f:
    event.ParseFromString(f.read())
    return load_tensor_from_event(event) 

def add_summary(self, summary, global_step=None):
    """Adds a `Summary` protocol buffer to the event file.

    This method wraps the provided summary in an `Event` protocol buffer
    and adds it to the event file.

    You can pass the result of evaluating any summary op, using
    @{tf.Session.run} or
    @{tf.Tensor.eval}, to this
    function. Alternatively, you can pass a `tf.Summary` protocol
    buffer that you populate with your own data. The latter is
    commonly done to report evaluation results in event files.

    Args:
      summary: A `Summary` protocol buffer, optionally serialized as a string.
      global_step: Number. Optional global step value to record with the
        summary.
    """
    if isinstance(summary, bytes):
      summ = summary_pb2.Summary()
      summ.ParseFromString(summary)
      summary = summ
    event = event_pb2.Event(summary=summary)
    self._add_event(event, global_step) 

def add_meta_graph(self, meta_graph_def, global_step=None):
    """Adds a `MetaGraphDef` to the event file.

    The `MetaGraphDef` allows running the given graph via
    `saver.import_meta_graph()`.

    Args:
      meta_graph_def: A `MetaGraphDef` object, often as returned by
        `saver.export_meta_graph()`.
      global_step: Number. Optional global step counter to record with the
        graph.

    Raises:
      TypeError: If both `meta_graph_def` is not an instance of `MetaGraphDef`.
    """
    if not isinstance(meta_graph_def, meta_graph_pb2.MetaGraphDef):
      raise TypeError("meta_graph_def must be type MetaGraphDef, saw type: %s" %
                      type(meta_graph_def))
    meta_graph_bytes = meta_graph_def.SerializeToString()
    event = event_pb2.Event(meta_graph_def=meta_graph_bytes)
    self._add_event(event, global_step) 

def Load(self):
    """Loads all new values from disk.

    Calling Load multiple times in a row will not 'drop' events as long as the
    return value is not iterated over.

    Yields:
      All values that were written to disk that have not been yielded yet.
    """
    while True:
      try:
        with errors.raise_exception_on_not_ok_status() as status:
          self._reader.GetNext(status)
      except (errors.DataLossError, errors.OutOfRangeError):
        # We ignore partial read exceptions, because a record may be truncated.
        # PyRecordReader holds the offset prior to the failed read, so retrying
        # will succeed.
        break
      event = event_pb2.Event()
      event.ParseFromString(self._reader.record())
      yield event
    logging.debug('No more events in %s', self._file_path) 

def add_summary(self, summary, global_step=None):
    """Adds a `Summary` protocol buffer to the event file.

    This method wraps the provided summary in an `Event` protocol buffer
    and adds it to the event file.

    You can pass the result of evaluating any summary op, using
    [`Session.run()`](client.md#Session.run) or
    [`Tensor.eval()`](framework.md#Tensor.eval), to this
    function. Alternatively, you can pass a `tf.Summary` protocol
    buffer that you populate with your own data. The latter is
    commonly done to report evaluation results in event files.

    Args:
      summary: A `Summary` protocol buffer, optionally serialized as a string.
      global_step: Number. Optional global step value to record with the
        summary.
    """
    if isinstance(summary, bytes):
      summ = summary_pb2.Summary()
      summ.ParseFromString(summary)
      summary = summ
    event = event_pb2.Event(summary=summary)
    self._add_event(event, global_step) 

def add_run_metadata(self, run_metadata, tag, global_step=None):
    """Adds a metadata information for a single session.run() call.

    Args:
      run_metadata: A `RunMetadata` protobuf object.
      tag: The tag name for this metadata.
      global_step: Number. Optional global step counter to record with the
        StepStats.

    Raises:
      ValueError: If the provided tag was already used for this type of event.
    """
    if tag in self._session_run_tags:
      raise ValueError("The provided tag was already used for this event type")
    self._session_run_tags[tag] = True

    tagged_metadata = event_pb2.TaggedRunMetadata()
    tagged_metadata.tag = tag
    # Store the `RunMetadata` object as bytes in order to have postponed
    # (lazy) deserialization when used later.
    tagged_metadata.run_metadata = run_metadata.SerializeToString()
    event = event_pb2.Event(tagged_run_metadata=tagged_metadata)
    self._add_event(event, global_step) 

def Load(self):
    """Loads all new values from disk.

    Calling Load multiple times in a row will not 'drop' events as long as the
    return value is not iterated over.

    Yields:
      All values that were written to disk that have not been yielded yet.
    """
    while True:
      try:
        with errors.raise_exception_on_not_ok_status() as status:
          self._reader.GetNext(status)
      except (errors.DataLossError, errors.OutOfRangeError):
        # We ignore partial read exceptions, because a record may be truncated.
        # PyRecordReader holds the offset prior to the failed read, so retrying
        # will succeed.
        break
      event = event_pb2.Event()
      event.ParseFromString(self._reader.record())
      yield event
    logging.debug('No more events in %s', self._file_path) 

def event_to_record(event):
    """
    Convert an event protobuf to a TFRecord

    Arguments:
        event (Event): Event Protobuf to write in TFRecord format

    Returns:
        TFRecord formatted bytestring
    """

    event_str = serialize_protobuf(event)
    header = struct.pack('Q', len(event_str))
    record = [header,
              struct.pack('I', masked_crc32c(header)),
              event_str,
              struct.pack('I', masked_crc32c(event_str))]

    return b"".join(record) 

def add_meta_graph(self, meta_graph_def, global_step=None):
    """Adds a `MetaGraphDef` to the event file.

    The `MetaGraphDef` allows running the given graph via
    `saver.import_meta_graph()`.

    Args:
      meta_graph_def: A `MetaGraphDef` object, often as retured by
        `saver.export_meta_graph()`.
      global_step: Number. Optional global step counter to record with the
        graph.

    Raises:
      TypeError: If both `meta_graph_def` is not an instance of `MetaGraphDef`.
    """
    if not isinstance(meta_graph_def, meta_graph_pb2.MetaGraphDef):
      raise TypeError("meta_graph_def must be type MetaGraphDef, saw type: %s"
                      % type(meta_graph_def))
    meta_graph_bytes = meta_graph_def.SerializeToString()
    event = event_pb2.Event(meta_graph_def=meta_graph_bytes)
    self._add_event(event, global_step) 

def add_run_metadata(self, run_metadata, tag, global_step=None):
    """Adds a metadata information for a single session.run() call.

    Args:
      run_metadata: A `RunMetadata` protobuf object.
      tag: The tag name for this metadata.
      global_step: Number. Optional global step counter to record with the
        StepStats.

    Raises:
      ValueError: If the provided tag was already used for this type of event.
    """
    if tag in self._session_run_tags:
      raise ValueError("The provided tag was already used for this event type")
    self._session_run_tags[tag] = True

    tagged_metadata = event_pb2.TaggedRunMetadata()
    tagged_metadata.tag = tag
    # Store the `RunMetadata` object as bytes in order to have postponed
    # (lazy) deserialization when used later.
    tagged_metadata.run_metadata = run_metadata.SerializeToString()
    event = event_pb2.Event(tagged_run_metadata=tagged_metadata)
    self._add_event(event, global_step) 

def log_key_value(self, key, value, step=None):
        def summary_val(k, v):
            kwargs = {'tag': k, 'simple_value': float(v)}
            return tf.Summary.Value(**kwargs)

        summary = tf.Summary(value=[summary_val(key, value)])
        event = event_pb2.Event(wall_time=time.time(), summary=summary)
        # Use a separate step counter for each key
        if key not in self.key_steps:
            self.key_steps[key] = 0
        if step is not None:
            self.key_steps[key] = step
        event.step = self.key_steps[key]
        self.writer.WriteEvent(event)
        self.writer.Flush()
        self.key_steps[key] += 1 

def __init__(self, logdir, flush_secs=2, is_dummy=False, dummy_time=None):
        self._name_to_tf_name = {}
        self._tf_names = set()
        self.is_dummy = is_dummy
        self.logdir = logdir
        self.flush_secs = flush_secs  # TODO
        self._writer = None
        self._dummy_time = dummy_time
        if is_dummy:
            self.dummy_log = defaultdict(list)
        else:
            if not os.path.exists(self.logdir):
                os.makedirs(self.logdir)
            hostname = socket.gethostname()
            filename = os.path.join(
                self.logdir, 'events.out.tfevents.{}.{}'.format(
                    int(self._time()), hostname))
            self._writer = open(filename, 'wb')
            self._write_event(event_pb2.Event(
                wall_time=self._time(), step=0, file_version='brain.Event:2')) 

def load_tensor_from_event_file(event_file_path):
  """Load a tensor from an event file.

  Assumes that the event file contains a `Event` protobuf and the `Event`
  protobuf contains a `Tensor` value.

  Args:
    event_file_path: (`str`) path to the event file.

  Returns:
    The tensor value loaded from the event file, as a `numpy.ndarray`. For
    uninitialized Tensors, returns `None`. For Tensors of data types that
    cannot be converted to `numpy.ndarray` (e.g., `tf.resource`), return
    `None`.
  """

  event = event_pb2.Event()
  with gfile.Open(event_file_path, "rb") as f:
    event.ParseFromString(f.read())
    return load_tensor_from_event(event) 

def write_values(self, key2val):
        summary = tf.Summary(value=[tf.Summary.Value(tag=k, simple_value=float(v))
            for (k, v) in key2val.items()])
        event = event_pb2.Event(wall_time=time.time(), summary=summary)
        event.step = self.step # is there any reason why you'd want to specify the step?
        self.evwriter.WriteEvent(event)
        self.evwriter.Flush()
        self.step += 1 

def load_tensor_from_event(event):
  """Load a tensor from an Event proto.

  Args:
    event: The Event proto, assumed to hold a tensor value in its
        summary.value[0] field.

  Returns:
    The tensor value loaded from the event file, as a `numpy.ndarray`, if
    representation of the tensor value by a `numpy.ndarray` is possible.
    For uninitialized Tensors, returns `None`. For Tensors of data types that
    cannot be represented as `numpy.ndarray` (e.g., `tf.resource`), return
    the `TensorProto` protobuf object without converting it to a
    `numpy.ndarray`.
  """

  tensor_proto = event.summary.value[0].tensor
  if tensor_proto.tensor_content or tensor_proto.string_val:
    # Initialized tensor.
    if tensor_proto.dtype == types_pb2.DT_RESOURCE:
      tensor_value = InconvertibleTensorProto(tensor_proto)
    else:
      try:
        tensor_value = tensor_util.MakeNdarray(tensor_proto)
      except KeyError:
        tensor_value = InconvertibleTensorProto(tensor_proto)
  else:
    # Uninitialized tensor or tensor of unconvertible data type.
    tensor_value = InconvertibleTensorProto(tensor_proto, False)

  return tensor_value 

def _load_graph_def_from_event_file(event_file_path):
  event = event_pb2.Event()
  with gfile.Open(event_file_path, "rb") as f:
    event.ParseFromString(f.read())

  return graph_pb2.GraphDef.FromString(event.graph_def) 

def _load_log_message_from_event_file(event_file_path):
  event = event_pb2.Event()
  with gfile.Open(event_file_path, "rb") as f:
    event.ParseFromString(f.read())

  return event.log_message.message 

def get_tensor(self):
    """Get tensor from the dump (`Event`) file.

    Returns:
      The tensor loaded from the dump (`Event`) file.
    """

    return load_tensor_from_event_file(self.file_path)

  # TODO(cais): Add time unit suffix to timestamp and t0 (us). 

def WriteEvent(self, event):
      from tensorflow.core.util.event_pb2 import Event
      if not isinstance(event, Event):
        raise TypeError("Expected an event_pb2.Event proto, "
                        " but got %s" % type(event))
      return self._WriteSerializedEvent(event.SerializeToString()) 

def log(level, message, *args):
  """Conditionally logs `message % args` at the level `level`.

  Note that tensorboard_logging verbosity and logging verbosity are separate;
  the message will always be passed through to the logging module regardless of
  whether it passes the tensorboard_logging verbosity check.

  Args:
    level: The verbosity level to use. Must be one of
      tensorboard_logging.{DEBUG, INFO, WARN, ERROR, FATAL}.
    message: The message template to use.
    *args: Arguments to interpolate to the message template, if any.

  Raises:
    ValueError: If `level` is not a valid logging level.
    RuntimeError: If the `SummaryWriter` to use has not been set.
  """
  if _summary_writer is _sentinel_summary_writer:
    raise RuntimeError('Must call set_summary_writer before doing any '
                       'logging from tensorboard_logging')
  _check_verbosity(level)
  proto_level = _LEVEL_PROTO_MAP[level]
  if proto_level >= _LEVEL_PROTO_MAP[_verbosity]:
    log_message = event_pb2.LogMessage(level=proto_level,
                                       message=message % args)
    event = event_pb2.Event(wall_time=time.time(), log_message=log_message)

    if _summary_writer:
      _summary_writer.add_event(event)

  logging.log(_PLATFORM_LOGGING_LEVEL_MAP[level], message, *args) 

def __init__(self, logdir, max_queue=10, flush_secs=120,
               filename_suffix=None):
    """Creates a `EventFileWriter` and an event file to write to.

    On construction the summary writer creates a new event file in `logdir`.
    This event file will contain `Event` protocol buffers, which are written to
    disk via the add_event method.

    The other arguments to the constructor control the asynchronous writes to
    the event file:

    *  `flush_secs`: How often, in seconds, to flush the added summaries
       and events to disk.
    *  `max_queue`: Maximum number of summaries or events pending to be
       written to disk before one of the 'add' calls block.

    Args:
      logdir: A string. Directory where event file will be written.
      max_queue: Integer. Size of the queue for pending events and summaries.
      flush_secs: Number. How often, in seconds, to flush the
        pending events and summaries to disk.
      filename_suffix: A string. Every event file's name is suffixed with
        `filename_suffix`.
    """
    self._logdir = logdir
    if not gfile.IsDirectory(self._logdir):
      gfile.MakeDirs(self._logdir)
    self._event_queue = six.moves.queue.Queue(max_queue)
    self._ev_writer = pywrap_tensorflow.EventsWriter(
        compat.as_bytes(os.path.join(self._logdir, "events")))
    self._flush_secs = flush_secs
    self._sentinel_event = self._get_sentinel_event()
    if filename_suffix:
      self._ev_writer.InitWithSuffix(compat.as_bytes(filename_suffix))
    self._closed = False
    self._worker = _EventLoggerThread(self._event_queue, self._ev_writer,
                                      self._flush_secs, self._sentinel_event)

    self._worker.start() 

def _get_sentinel_event(self):
    """Generate a sentinel event for terminating worker."""
    return event_pb2.Event() 

def add_session_log(self, session_log, global_step=None):
    """Adds a `SessionLog` protocol buffer to the event file.

    This method wraps the provided session in an `Event` protocol buffer
    and adds it to the event file.

    Args:
      session_log: A `SessionLog` protocol buffer.
      global_step: Number. Optional global step value to record with the
        summary.
    """
    event = event_pb2.Event(session_log=session_log)
    self._add_event(event, global_step) 

def _add_graph_def(self, graph_def, global_step=None):
    graph_bytes = graph_def.SerializeToString()
    event = event_pb2.Event(graph_def=graph_bytes)
    self._add_event(event, global_step) 

def add_event(self, event):
    """Adds an event to the event file.

    Args:
      event: An `Event` protocol buffer.
    """
    self.event_writer.add_event(event) 

def add_summary(self, summary, global_step=None):
    """Adds a `Summary` protocol buffer to the event file.

    This method wraps the provided summary in an `Event` protocol buffer
    and adds it to the event file.

    You can pass the result of evaluating any summary op, using
    @{tf.Session.run} or
    @{tf.Tensor.eval}, to this
    function. Alternatively, you can pass a `tf.Summary` protocol
    buffer that you populate with your own data. The latter is
    commonly done to report evaluation results in event files.

    Args:
      summary: A `Summary` protocol buffer, optionally serialized as a string.
      global_step: Number. Optional global step value to record with the
        summary.
    """
    if isinstance(summary, bytes):
      summ = summary_pb2.Summary()
      summ.ParseFromString(summary)
      summary = summ
    event = event_pb2.Event(wall_time=time.time(), summary=summary)
    if global_step is not None:
      event.step = int(global_step)
    self.add_event(event) 

def add_session_log(self, session_log, global_step=None):
    """Adds a `SessionLog` protocol buffer to the event file.

    This method wraps the provided session in an `Event` protocol buffer
    and adds it to the event file.

    Args:
      session_log: A `SessionLog` protocol buffer.
      global_step: Number. Optional global step value to record with the
        summary.
    """
    event = event_pb2.Event(wall_time=time.time(), session_log=session_log)
    if global_step is not None:
      event.step = int(global_step)
    self.add_event(event) 

def _add_graph_def(self, graph_def, global_step=None):
    graph_bytes = graph_def.SerializeToString()
    event = event_pb2.Event(wall_time=time.time(), graph_def=graph_bytes)
    if global_step is not None:
      event.step = int(global_step)
    self._event_queue.put(event) 

def add_run_metadata(self, run_metadata, tag, global_step=None):
    """Adds a metadata information for a single session.run() call.

    Args:
      run_metadata: A `RunMetadata` protobuf object.
      tag: The tag name for this metadata.
      global_step: Number. Optional global step counter to record with the
        StepStats.

    Raises:
      ValueError: If the provided tag was already used for this type of event.
    """
    if tag in self._session_run_tags:
      raise ValueError("The provided tag was already used for this event type")
    self._session_run_tags[tag] = True

    tagged_metadata = event_pb2.TaggedRunMetadata()
    tagged_metadata.tag = tag
    # Store the `RunMetadata` object as bytes in order to have postponed
    # (lazy) deserialization when used later.
    tagged_metadata.run_metadata = run_metadata.SerializeToString()
    event = event_pb2.Event(wall_time=time.time(),
                            tagged_run_metadata=tagged_metadata)
    if global_step is not None:
      event.step = int(global_step)
    self._event_queue.put(event)