Python Engine.get_train_start_epoch_batch方法代码示例

# 需要导入模块: from Engine import Engine [as 别名]
# 或者: from Engine.Engine import get_train_start_epoch_batch [as 别名]
def iterateEpochs():
  start_epoch, start_batch = Engine.get_train_start_epoch_batch(config)
  final_epoch = Engine.config_get_final_epoch(config)

  print("Starting with epoch %i, batch %i." % (start_epoch, start_batch), file=log.v3)
  print("Final epoch is: %i" % final_epoch, file=log.v3)

  recurrent_net = "lstm" in config.value("hidden_type", "")  # good enough...
  batch_size = config.int('batch_size', 1)
  max_seqs = config.int('max_seqs', -1)

  for epoch in range(start_epoch, final_epoch + 1):
    print("Epoch %i." % epoch, file=log.v3)
    rnn.train_data.init_seq_order(epoch)
    iterateDataset(rnn.train_data, recurrent_net=recurrent_net, batch_size=batch_size, max_seqs=max_seqs)

  print("Finished all epochs.", file=log.v3)

# 需要导入模块: from Engine import Engine [as 别名]
# 或者: from Engine.Engine import get_train_start_epoch_batch [as 别名]
def getSegmentList(corpusName, segmentList, **kwargs):
  """
  Called by Sprint PythonSegmentOrder.
  Set python-segment-order = true in Sprint to use this.

  If this is used, this gets called really early.
  If it is used together with the Sprint PythonTrainer,
  it will get called way earlier before the init() below.
  It might also get called multiple times, e.g. if
  Sprint is in interactive mode to calc the seg count.
  This is optional. You can use the SprintInterface
  only for the PythonTrainer.

  :type corpusName: str
  :type segmentList: list[str]
  :type segmentsInfo: dict[str,dict[str]]
  :rtype: list[str]
  :returns segment list. Can also be an iterator.
  """
  print("Sprint: getSegmentList(%r)" % corpusName)
  print("Corpus segments #: %i" % len(segmentList))
  print("(This can be further filtered in Sprint by a whitelist or so.)")

  # Init what we need. These can be called multiple times.
  # If we use both the PythonSegmentOrder and the PythonTrainer, this will be called first.
  # The PythonTrainer will be called lazily once it gets the first data.
  initBase()
  sprintDataset.useMultipleEpochs()

  finalEpoch = getFinalEpoch()
  startEpoch, startSegmentIdx = Engine.get_train_start_epoch_batch(config)
  print("Sprint: Starting with epoch %i, segment-idx %s." % (startEpoch, startSegmentIdx))
  print("Final epoch is: %i" % finalEpoch)

  # Loop over multiple epochs. Epochs start at 1.
  for curEpoch in range(startEpoch, finalEpoch + 1):
    if isTrainThreadStarted:
      # So that the CRNN train thread always has the SprintDataset in a sane state before we reset it.
      sprintDataset.waitForCrnnEpoch(curEpoch)
    sprintDataset.initSprintEpoch(curEpoch)

    index_list = sprintDataset.get_seq_order_for_epoch(curEpoch, len(segmentList))
    orderedSegmentList = [segmentList[i] for i in index_list]
    assert len(orderedSegmentList) == len(segmentList)

    print("Sprint epoch: %i" % curEpoch)
    startSegmentIdx = 0
    if curEpoch == startEpoch: startSegmentIdx = startSegmentIdx
    for curSegmentIdx in range(startSegmentIdx, len(orderedSegmentList)):
      sprintDataset.set_complete_frac(float(curSegmentIdx - startSegmentIdx + 1) /
                                      (len(orderedSegmentList) - startSegmentIdx))
      yield orderedSegmentList[curSegmentIdx]

    print("Sprint finished epoch %i" % curEpoch)
    sprintDataset.finishSprintEpoch()

    if isTrainThreadStarted:
      assert sprintDataset.get_num_timesteps() > 0, \
        "We did not received any seqs. You are probably using a buffered feature extractor and the buffer is " + \
        "bigger than the total number of time frames in the corpus."

  sprintDataset.finalizeSprint()