Reputation: 41
I want to convert csv format dataset files to TFRecords parallelly. Is tf.python_io.TFRecordWriter
thread-safe?
Upvotes: 4
Views: 608
Reputation: 125
I was also searching for the question. I found at it is not a thread safe method. I couldn't find any documentation for this answer but tried both multi threading process and single threading process instead on same data. The result was that multi threading approach creates a record file that throws an error while training and single threading approach creates a valid record file which of course runs fine while training but takes much longer time than multi threading during creation time. I hope this answer is helpful.
Upvotes: 1