In working with Tensorflow and TFLearn on Windows I frequently run into a problem with my source data files being encoded as UTF-8 with a BOM header. A BOM is a byte order mark, a single unicode character that prefaces the file. Many data loading utilities load with the incorrect encoding and will throw a
ValueError about this unexpected character.
The large files often used as training data can be challenging to open/re-encode properly so I created this method to rewrite the file in-place without the BOM mark.
import os, sys, codecs def remove_bom_inplace(path): """Removes BOM mark, if it exists, from a file and rewrites it in-place""" buffer_size = 4096 bom_length = len(codecs.BOM_UTF8) with open(path, "r+b") as fp: chunk = fp.read(buffer_size) if chunk.startswith(codecs.BOM_UTF8): i = 0 chunk = chunk[bom_length:] while chunk: fp.seek(i) fp.write(chunk) i += len(chunk) fp.seek(bom_length, os.SEEK_CUR) chunk = fp.read(buffer_size) fp.seek(-bom_length, os.SEEK_CUR) fp.truncate()
I actually use this in my TFLearn training scripts automatically, like this:
try: data, labels = load_csv(...) except ValueError: strip_bom.remove_bom_inplace(filename) data, labels = load_csv(...)
Such that any BOM related error during loading attempts a repair of the file, and a retry automatically.