tfrecord下载tfrecord源代码下载

Tfrecord读者和作家

该库允许在Python中有效地读取和编写Tfrecord文件。该库还为Pytorch提供了TfreCord文件的iTerabledataSet读取器。目前支持未压缩和压缩的Gzip Tfrecords。

安装

 pip3 install 'tfrecord[torch]'

用法

建议为每个Tfrecord文件创建一个索引文件。使用多个工人时必须提供索引文件，否则加载程序可以返回重复记录。您可以使用此实用程序程序为单个tfrecord文件创建索引文件：

 python3 -m tfrecord.tools.tfrecord2idx <tfrecord path> <index path>

在目录运行中创建所有“ .tfidnex”文件的“ .tfidnex”文件：

 tfrecord2idx <data dir>

阅读和写作tf.train.example

读取tf。示例记录pytorch

使用tfrecorddataset读取pytorch中的tfrecord文件。

 import torch
from tfrecord . torch . dataset import TFRecordDataset

tfrecord_path = "/tmp/data.tfrecord"
index_path = None
description = { "image" : "byte" , "label" : "float" }
dataset = TFRecordDataset ( tfrecord_path , index_path , description )
loader = torch . utils . data . DataLoader ( dataset , batch_size = 32 )

data = next ( iter ( loader ))
print ( data )

使用MultitFrecordDataSet读取多个Tfrecord文件。此类从给定的tfrecord文件中的示例带有给定的概率。

 import torch
from tfrecord . torch . dataset import MultiTFRecordDataset

tfrecord_pattern = "/tmp/{}.tfrecord"
index_pattern = "/tmp/{}.index"
splits = {
    "dataset1" : 0.8 ,
    "dataset2" : 0.2 ,
}
description = { "image" : "byte" , "label" : "int" }
dataset = MultiTFRecordDataset ( tfrecord_pattern , index_pattern , splits , description )
loader = torch . utils . data . DataLoader ( dataset , batch_size = 32 )

data = next ( iter ( loader ))
print ( data )

无限和有限的Pytorch数据集

默认情况下， MultiTFRecordDataset是无限的，这意味着它将永远采样数据。您可以通过提供适当的标志来使其有限

 dataset = MultiTFRecordDataset(..., infinite=False)

改组数据

当您提供队列尺寸时，TfrecordDataSet和MultitFrecordDataSet都会自动将数据供电。

 dataset = TFRecordDataset(..., shuffle_queue_size=1024)

转换输入数据

您可以选择将函数作为transform参数传递，以在返回之前执行功能的后处理。例如，这可以用于解码图像或将颜色标准化为特定范围或垫变量长度序列。

 import tfrecord
import cv2

def decode_image ( features ):
    # get BGR image from bytes
    features [ "image" ] = cv2 . imdecode ( features [ "image" ], - 1 )
    return features


description = {
    "image" : "bytes" ,
}

dataset = tfrecord . torch . TFRecordDataset ( "/tmp/data.tfrecord" ,
                                         index_path = None ,
                                         description = description ,
                                         transform = decode_image )

data = next ( iter ( dataset ))
print ( data )

在python中编写tf.xample记录

 import tfrecord

writer = tfrecord . TFRecordWriter ( "/tmp/data.tfrecord" )
writer . write ({
    "image" : ( image_bytes , "byte" ),
    "label" : ( label , "float" ),
    "index" : ( index , "int" )
})
writer . close ()

读取python中的tf。示例记录

 import tfrecord

loader = tfrecord . tfrecord_loader ( "/tmp/data.tfrecord" , None , {
    "image" : "byte" ,
    "label" : "float" ,
    "index" : "int"
})
for record in loader :
    print ( record [ "label" ])

阅读和写作tf.train.sequenceexample

可以使用上面显示的相同方法读取和编写sequenceExamples，并带有额外的参数（用于读取的sequence_description和sequence_datum用于写作），从而导致相应的读/写功能将数据视为sequenceExample。

编写序列化示例以归档

 import tfrecord

writer = tfrecord . TFRecordWriter ( "/tmp/data.tfrecord" )
writer . write ({ 'length' : ( 3 , 'int' ), 'label' : ( 1 , 'int' )},
             { 'tokens' : ([[ 0 , 0 , 1 ], [ 0 , 1 , 0 ], [ 1 , 0 , 0 ]], 'int' ), 'seq_labels' : ([ 0 , 1 , 1 ], 'int' )})
writer . write ({ 'length' : ( 3 , 'int' ), 'label' : ( 1 , 'int' )},
             { 'tokens' : ([[ 0 , 0 , 1 ], [ 1 , 0 , 0 ]], 'int' ), 'seq_labels' : ([ 0 , 1 ], 'int' )})
writer . close ()

在Python中读取sequenceexamples

从序列样本yeilds读取一个包含两个元素的元组。

 import tfrecord

context_description = { "length" : "int" , "label" : "int" }
sequence_description = { "tokens" : "int" , "seq_labels" : "int" }
loader = tfrecord . tfrecord_loader ( "/tmp/data.tfrecord" , None ,
                                  context_description ,
                                  sequence_description = sequence_description )

for context , sequence_feats in loader :
    print ( context [ "label" ])
    print ( sequence_feats [ "seq_labels" ])

阅读pytorch中的序列示例

如有关Transforming Input的部分所述，可以将函数作为transform参数传递，以执行特征的后处理。对于序列特征，应该使用它，因为这些是可变的长度序列，并且需要在批处理之前填充。

 import torch
import numpy as np
from tfrecord . torch . dataset import TFRecordDataset

PAD_WIDTH = 5
def pad_sequence_feats ( data ):
    context , features = data
    for k , v in features . items ():
        features [ k ] = np . pad ( v , (( 0 , PAD_WIDTH - len ( v )), ( 0 , 0 )), 'constant' )
    return ( context , features )

context_description = { "length" : "int" , "label" : "int" }
sequence_description = { "tokens" : "int " , "seq_labels" : "int" }
dataset = TFRecordDataset ( "/tmp/data.tfrecord" ,
                          index_path = None ,
			  description = context_description ,
			  transform = pad_sequence_feats ,
			  sequence_description = sequence_description )
loader = torch . utils . data . DataLoader ( dataset , batch_size = 32 )
data = next ( iter ( loader ))
print ( data )

另外，您可以选择实现自定义collate_fn ，以组装批处理，例如执行动态填充。

 import torch
import numpy as np
from tfrecord . torch . dataset import TFRecordDataset

def collate_fn ( batch ):
    from torch . utils . data . _utils import collate
    from torch . nn . utils import rnn
    context , feats = zip ( * batch )
    feats_ = { k : [ torch . Tensor ( d [ k ]) for d in feats ] for k in feats [ 0 ]}
    return ( collate . default_collate ( context ),
            { k : rnn . pad_sequence ( f , True ) for ( k , f ) in feats_ . items ()})

context_description = { "length" : "int" , "label" : "int" }
sequence_description = { "tokens" : "int " , "seq_labels" : "int" }
dataset = TFRecordDataset ( "/tmp/data.tfrecord" ,
                          index_path = None ,
			  description = context_description ,
			  transform = pad_sequence_feats ,
			  sequence_description = sequence_description )
loader = torch . utils . data . DataLoader ( dataset , batch_size = 32 , collate_fn = collate_fn )
data = next ( iter ( loader ))
print ( data )