Improve usability of AutoML column not found error

Let's make the error message more actionable.

Error user sees:

I would recommend adding similar named column(s):

- $"Provided {columnPurpose} column '{columnName}' not found in training data."
+ $"Provided {columnPurpose} column '{columnName}' not found in training data. Did you mean '{closestNamed}'."

For my current example, this would print: Provided ignored column 'tagMaxTotalItem' not found in training data. Did you mean 'tagMaxTotalItems'.

I'd recommend using Levenshtein distance to find the closest named column (code).

Code location:

machinelearning/src/Microsoft.ML.AutoML/Utils/UserInputValidationUtil.cs

Lines 248 to 252 in 5dbfd8a

    
           var nullableColumn = trainData.Schema.GetColumnOrNull(columnName); 
        
           if (nullableColumn == null) 
        
           { 
        
               throw new ArgumentException($"Provided {columnPurpose} column '{columnName}' not found in training data."); 
        
           }

Background:
It took me ~20min to debug why this error was occurring (obvious in retrospect). My column existed in the dataset, it existed in my loader function, it existed in my IDataView, ...; simply was just misspelt ("tagMaxTotalItem" instead of "tagMaxTotalItems").

Improving the usability of this error message will save future users' time.

Offer suggestions for possibly mistyped label column names in AutoML (#…

Merge from main repository (#1)

Actually merge from main (#2)

hi
Is this issue solved ??

I was going to try working on this as my first contribution to open source, but it looks already fixed to me in the code. I don't see a test for it though. I could add a test to confirm it. I haven't seen any guidelines on writing tests. Are there any?

I dont think we have any official guidelines. I would just look at the other tests we have for AutoML and base it off of that. If you can throw the test in to an existing class that would probably be better, but if it doesn't fit anywhere than feel free to make a new test class too.

We can use sequence matcher ,it compare the column typed by us with the existing columns like this-
from difflib import SequenceMatcher
s = SequenceMatcher(None, a, b)
s.ratio()
if we find s.ratio() >50 than we will print
throw new ArgumentException($"Provided {columnPurpose} column '{columnName}' not found in training data. And their may be spelling mistake typed by you");
so it will help us in identifying such type of bugs.

Update editor config and format samples (#5893)

Actually what we have to do now? I am new to it plz help me and I really want to contribute on any issue .

Hi, I wanted to try and fix this issue but it seems this issue is already fixed:

machinelearning/src/Microsoft.ML.AutoML/Utils/UserInputValidationUtil.cs

Lines 254 to 258 in 4bcb892

    
           var exceptionMessage = $"Provided {columnPurpose} column '{columnName}' not found in training data."; 
        
           if (closestNamed != string.Empty) 
        
           { 
        
               exceptionMessage += $" Did you mean '{closestNamed}'."; 
        
           }

Can someone close this?

@the-programmer

Thanks for pointing this out @the-programmer! Closing the issue.

Navigation Menu

Repository navigation

Improve usability of AutoML column not found error #5574

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	var nullableColumn = trainData.Schema.GetColumnOrNull(columnName);
	if (nullableColumn == null)
	{
	throw new ArgumentException($"Provided {columnPurpose} column '{columnName}' not found in training data.");
	}

	var exceptionMessage = $"Provided {columnPurpose} column '{columnName}' not found in training data.";
	if (closestNamed != string.Empty)
	{
	exceptionMessage += $" Did you mean '{closestNamed}'.";
	}

Improve usability of AutoML column not found error #5574

Description

Activity

behroozbc commented on Jul 24, 2021

woodya-coder commented on Aug 2, 2021

michaelgsharp commented on Aug 4, 2021

abhinavjainn412 commented on Aug 21, 2021

bhargavsonagara commented on Aug 25, 2021

the-programmer commented on May 22, 2022

michaelgsharp commented on Jun 13, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions