System Total Fail due to double-Number conversion problems for non "es-US" Cultures

[andy-soft]

Hi, I trained the sample, for english BIO NER labeler, adn the training was OK But when running the TEST batch, it fails TOTALLY! The labels got were absolutely NUTS!

so I started to din inside algorithms, and found a BIG PROBLEM when you write down the training and model files, and use a "string" as double, in "en-US" culture, the decimal point is a "." while in spanish culture, it is a "," so when writing down and reading agfain the files, there is a inconsistency (if used as string) so the solution is 2 ways

1. use this way (modelReader.cs)

            //读入cost_factor
            strLine = sr.ReadLine();
            cost_factor_ = double.Parse(strLine.Split(':')[1].Trim(), CultureInfo.InvariantCulture);
   

2. add this to the first run inside a MAIN() loop for console apps.

   static void Main(string[] args)
    {
    	CultureInfo culture = CultureInfo.CreateSpecificCulture("en-US");
    	CultureInfo.DefaultThreadCurrentCulture = culture;
    	CultureInfo.DefaultThreadCurrentUICulture = culture;
    	Thread.CurrentThread.CurrentCulture = culture;
    	Thread.CurrentThread.CurrentUICulture = culture;
     .....
   

And now it works!