What did I do wrong with parsing MNIST dataset with BinaryReader in C#?

Question

I'm parsing MNIST datasets in C# from: http://yann.lecun.com/exdb/mnist/

I'm trying to read the first Int32 from a binary file:

FileStream fileS = new FileStream(fileName, FileMode.Open, FileAccess.Read);
BinaryReader reader = new BinaryReader(fileS);
int magicNumber = reader.ReadInt32();

However, it gives me a non-sense number: 50855936.

If I use File.ReadAllBytes()

buffer = File.ReadAllBytes(fileName);

and then look through the bytes, it works fine(the first four bytes now represents 2049), what did I do wrong with BinaryReader?

The file format is as follows (I'm trying to read the first magic number):

All the integers in the files are stored in the MSB first (high endian) format used by most non-Intel processors. Users of Intel processors and other low-endian machines must flip the bytes of the header.

TRAINING SET LABEL FILE (train-labels-idx1-ubyte):

[offset] [type]          [value]          [description] 
0000     32 bit integer  0x00000801(2049) magic number (MSB first) 
0004     32 bit integer  60000            number of items 
0008     unsignebyte     ??               label 
0009     unsigned byte   ??               label 
........ 
xxxx     unsigned byte   ??               label
The labels values are 0 to 9.d

What encoding is your file in? By default, BinaryReader uses UTF-8. Try specifying it in the BinaryReader contructor call, if it's different from UTF-8. — nestedloop, Jan 07 '14 at 08:34
What does it return, what did you expect? Maybe a Problem with endianess. — Henrik, Jan 07 '14 at 08:35
Well, now the documentation for [`ReadInt32`](http://msdn.microsoft.com/en-us/library/system.io.binaryreader.readint32(v=vs.110).aspx) and what you've added should make it clear. "BinaryReader reads this data type in little-endian format." — Damien_The_Unbeliever, Jan 07 '14 at 08:44
This may help: http://stackoverflow.com/questions/8620885/c-sharp-binary-reader-in-big-endian — Henrik, Jan 07 '14 at 08:45

score 8 · Accepted Answer · answered Jan 07 '14 at 09:01

8

50855936 == 0x03080000. Or 0x00000803 when you reverse the bytes, required on almost any machine since little-endian has won the egg war. Close enough to 2049, no great idea what explains the offset of 2. Here's an extension method to help you read it:

  public static class BigEndianUtils {
      public static int ReadBigInt32(this BinaryReader br) {
          var bytes = br.ReadBytes(sizeof(Int32));
          if (BitConverter.IsLittleEndian) Array.Reverse(bytes);
          return BitConverter.ToInt32(bytes, 0);
      }
  }

Add additional methods if the file contains more field types, just substitute Int32 in the snippet.

answered Jan 07 '14 at 09:01

Hans Passant

922,412
146
1,693
2,536

The 0x03080000 value is probably due to some other error (the OP probably just made an error typing the number). If you use BinaryWriter to write the 2049 value, it produces 0x01080000 as expected using little-endian. – Luaan Jan 07 '14 at 09:10
For posteriority - OP is parsing **MNIST** database in C# (http://yann.lecun.com/exdb/mnist/) and **0x00000803** (2051) and **0x00000801** (2049) difference is due to mistake in copying over from mentioned site. 2049 stands for 'magic number' for **train-labels-idx1-ubyte** dataset and 2051 stands for 'magic number' for **train-images-idx3-ubyte** dataset from MNIST. – mwilczynski Jan 16 '18 at 20:33

Luaan · Answer 2 · 2014-01-07T09:05:15.113

It seems that your problem is somewhere else. Could you post a minimal compilable snippet that doesn't work as expected?

For example, this snippet works exactly as expected - it creates a binary file of 8 bytes, which are two big-endian Int32s. The reader then correctly reads the data as the two integers.

using (var str = File.Create("C:\\Test.dat"))
    using (var wr = new BinaryWriter(str))
    {
        wr.Write(2049);
        wr.Write(60000);
    }

using (var str = File.Open("C:\\Test.dat", FileMode.Open))
    using (var rdr = new BinaryReader(str))
    {
        rdr.ReadInt32().Dump();
        rdr.ReadInt32().Dump();
    }

However, the endianness is fixed. If you need to use MSB first, you need to read the bytes and convert them to integers yourself (or, you could of course invert the byte order using bitwise operations, if you're so inclined).

What did I do wrong with parsing MNIST dataset with BinaryReader in C#?

2 Answers2