Source code for tensorpack.predict.dataset

# -*- coding: utf-8 -*-
# File: dataset.py


import multiprocessing
import os
from abc import ABCMeta, abstractmethod
import six

from ..dataflow import DataFlow
from ..dataflow.remote import dump_dataflow_to_process_queue
from ..utils import logger
from ..utils.develop import HIDE_DOC
from ..utils.concurrency import DIE, OrderedResultGatherProc, ensure_proc_terminate
from ..utils.gpu import change_gpu, get_num_gpu
from ..utils.utils import get_tqdm
from .base import OfflinePredictor
from .concurrency import MultiProcessQueuePredictWorker
from .config import PredictConfig

__all__ = ['DatasetPredictorBase', 'SimpleDatasetPredictor',
           'MultiProcessDatasetPredictor']


[docs]@six.add_metaclass(ABCMeta) class DatasetPredictorBase(object): """ Base class for dataset predictors. These are predictors which run over a :class:`DataFlow`. """
[docs] def __init__(self, config, dataset): """ Args: config (PredictConfig): the config of predictor. dataset (DataFlow): the DataFlow to run on. """ assert isinstance(dataset, DataFlow) assert isinstance(config, PredictConfig) self.config = config self.dataset = dataset
[docs] @abstractmethod def get_result(self): """ Yields: output for each datapoint in the DataFlow. """ pass
[docs] def get_all_result(self): """ Returns: list: all outputs for all datapoints in the DataFlow. """ return list(self.get_result())
[docs]class SimpleDatasetPredictor(DatasetPredictorBase): """ Simply create one predictor and run it on the DataFlow. """ def __init__(self, config, dataset): super(SimpleDatasetPredictor, self).__init__(config, dataset) self.predictor = OfflinePredictor(config) @HIDE_DOC def get_result(self): self.dataset.reset_state() try: sz = len(self.dataset) except NotImplementedError: sz = 0 with get_tqdm(total=sz, disable=(sz == 0)) as pbar: for dp in self.dataset: res = self.predictor(*dp) yield res pbar.update()
[docs]class MultiProcessDatasetPredictor(DatasetPredictorBase): """ Run prediction in multiple processes, on either CPU or GPU. Each process fetch datapoints as tasks and run predictions independently. """ # TODO allow unordered
[docs] def __init__(self, config, dataset, nr_proc, use_gpu=True, ordered=True): """ Args: config: same as in :class:`DatasetPredictorBase`. dataset: same as in :class:`DatasetPredictorBase`. nr_proc (int): number of processes to use use_gpu (bool): use GPU or CPU. If GPU, then ``nr_proc`` cannot be more than what's in CUDA_VISIBLE_DEVICES. ordered (bool): produce outputs in the original order of the datapoints. This will be a bit slower. Otherwise, :meth:`get_result` will produce outputs in any order. """ if config.return_input: logger.warn("Using the option `return_input` in MultiProcessDatasetPredictor might be slow") assert nr_proc >= 1, nr_proc super(MultiProcessDatasetPredictor, self).__init__(config, dataset) self.nr_proc = nr_proc self.ordered = ordered self.inqueue, self.inqueue_proc = dump_dataflow_to_process_queue( self.dataset, nr_proc * 2, self.nr_proc) # put (idx, dp) to inqueue if use_gpu: try: gpus = os.environ['CUDA_VISIBLE_DEVICES'].split(',') except KeyError: gpus = list(range(get_num_gpu())) assert len(gpus) >= self.nr_proc, \ "nr_proc={} while only {} gpus available".format( self.nr_proc, len(gpus)) else: gpus = ['-1'] * self.nr_proc # worker produces (idx, result) to outqueue self.outqueue = multiprocessing.Queue() self.workers = [MultiProcessQueuePredictWorker( i, self.inqueue, self.outqueue, self.config) for i in range(self.nr_proc)] # start inqueue and workers self.inqueue_proc.start() for p, gpuid in zip(self.workers, gpus): if gpuid == '-1': logger.info("Worker {} uses CPU".format(p.idx)) else: logger.info("Worker {} uses GPU {}".format(p.idx, gpuid)) with change_gpu(gpuid): p.start() if ordered: self.result_queue = OrderedResultGatherProc( self.outqueue, nr_producer=self.nr_proc) self.result_queue.start() ensure_proc_terminate(self.result_queue) else: self.result_queue = self.outqueue ensure_proc_terminate(self.workers + [self.inqueue_proc])
@HIDE_DOC def get_result(self): try: sz = len(self.dataset) except NotImplementedError: sz = 0 with get_tqdm(total=sz, disable=(sz == 0)) as pbar: die_cnt = 0 while True: res = self.result_queue.get() pbar.update() if res[0] != DIE: yield res[1] else: die_cnt += 1 if die_cnt == self.nr_proc: break self.inqueue_proc.join() self.inqueue_proc.terminate() if self.ordered: # if ordered, than result_queue is a Process self.result_queue.join() self.result_queue.terminate() for p in self.workers: p.join() p.terminate()