Machinelearning: How to ignore a column

[rauhs]

System information

• OS version/distro: Win 10
• .NET Version (eg., dotnet --info): 2.1.201

Issue

• What did you do?
  I added a custom column in my input data type. It holds a DateTime. I need it for stats but not for learning. I never "copy" it to the "Features" column. I also added

new ColumnDropper() { Column = new []{ "PlannedStart"} }

• What happened?
  I got an ArgumentOutOfRangeException: Could not determine an IDataView type for member ...

• What did you expect?

I expected it to just ignore the column.

[MrCSharp22]

I think if you don't include the column in the "features" column then it is considered to be ignored.

Also, you can just not include a property for that column in the model that represents the CSV row.

[rauhs]

Unfortunately, that won't work. As soon as I add any other fields to the class ML.NET will want to do something with it. I solved the problem with deriving the class to a richer class which holds the fields I want. Seems to work

[TomFinley]

I'm not entirely certain what mechanism we're using to do this, but the field attributes seem to be declared in this file. One of the attributes available is the so-called NoColumn attribute, seen here:

https://github.com/dotnet/machinelearning/blob/828dc227f4d7346e11094479c7a2e443addc8102/src/Microsoft.ML.Api/SchemaDefinition.cs#L119-L123

From its description, it may do what you want.

[rauhs]

[NoColumn] works! Thanks Tom.

[asif48]

Hi Rauhs...
I have the same problem that I added a Date column in the input class

public class ItemStock
    {
        [Column("0")]
        public string CarID;

        [Column("1")]
        public float LocID;

        **[Column("2")]
        public DateTime Date;**

        [Column("3")]
        public float Punctured;
    }

I am having the error "Could not determine an IDataView type for member Date"...
Plz let me know how to resolve that...