Label-studio: sometime show remarked data when labeling.

[MarsSu0618]

Problems

Hi, I have encountered some problems when i label my data. The label studio tool sometimes show remarked data.
For example, If i had labeled the first data. but when i label 500 or more data, the first remarked data show again.
I have checked my data that do not duplicated. So I want to check if whether the tool bug?
Anyone have any ideas?

Environment

• label-studio: 0.8.0
• text data saved on aws s3.
• i used sentence label.

[makseq]

Hi, I think you use an old LS version. There are many bugfixes in 0.9, please, try it.

[MarsSu0618]

@makseq Thanks for your response.
I will try it with new version 0.9.0. But i want to check whether if the bug show in older version?

[makseq]

You need to remember task ID of doubled tasks, if they are the same it could be a bug.

[MarsSu0618]

@makseq Yes. Task ID sometimes duplicated. If i label task-id:101, it show 101 again after i label task-id:150
So upgrade version 0.9 that have this issue?

[makseq]

Can you check: when you see the same task again, has it a completion already? (Check it using task manager)

[MarsSu0618]

@makseq I have check it . the completion is already. But task id show again when i labeling.

[makseq]

Yep, I think it's a bug..

[MarsSu0618]

@makseq if this is a bug, it is fixed in version 0.9?

[MarsSu0618]

@makseq btw, I used version 0.9.0, connect aws s3 show some error message.follow as

[makseq]

In this case you have something wrong in s3 files. Are there are json files?

[MarsSu0618]

@makseq Yes, I have json file. format follow as:
json= [ {"text": "......"} ]
version 0.8.0 is normal, version 0.9.0 show error. and use same json file.

[makseq]

Yep, this is an error: you must not use list of tasks inside of your json files on s3. Only one task per file. It's a limitation by LS design.

[MarsSu0618]

so I need to remove list of task?

{"text": "......"}

and remark and task_id issue still not resolve, right? maybe cache?

[makseq]

yes, leave task as dict {} only.

and remark and task_id issue still not resolve

you mean duplicates in the labeling?

[MarsSu0618]

@makseq I consider the problem is User Interface bug.
Because I used version 0.9.0 UI that it do not have this problem. But used old user interface still have task_id duplicates problem.
So We decide to labeling with new interface. Thanks.

[MarsSu0618]

@makseq Hi, We found other problems about version 0.9.0.
Our data are save AWS S3. Then, Sometimes label studio interface scan have some problem. Please follow as photo.

It can not encode file text. Just show file link on s3. and current solution is restart tool service, it is normal.
Hope you can check what is problem. We guess maybe tool redirect and decode problem.

BTW.. new version user interface seem like do have RWD? right?

[makseq]

Sorry, what is RWD :slightly_smiling_face: ?

[makseq]

Just show file link on s3. and current solution is restart tool service, it is normal.

But what will you see if you go to the labeling? Does it load for labeling normally?

[MarsSu0618]

@makseq it can not how normal text. just show text json file link on s3.
Maybe issue like #497 .

[makseq]

What files do you store on your storage? txt or json? Can you show their examples?

[MarsSu0618]

@makseq it is json file.

data content:

{'text': '......'}

[makseq]

Ok, I'll check it by myself and come back to you.

[MarsSu0618]

@makseq Thank you.
I found you use ujson lib instead of json lib. Maybe it is problem?

[makseq]

As I know it's a full analogue of json. But much faster.

[MarsSu0618]

@makseq OK, Maybe you need check readable_path from s3.py....

[makseq]

If you have public s3 storage with these example files it will be much faster to check it for me.

[MarsSu0618]

@makseq
We found this problem maybe occurred in import interface. please follow as image:

When we click import button and show s3 error.
In addition, We found other solution, which reset data manager, and it become normal.

So, maybe you should check import function code.

[makseq]

@MarsSu0618 Hm, I see that you have a checkbox "Treat every bucket object as source file ...". In your case it must be off. Maybe something change its state unexpectedly?

[MarsSu0618]

@makseq I want to suggest about user interface of label studio. Hope can add search bar so that we can search marked data.

[makseq]

@MarsSu0618 why don't you use Filters over completion results?

[MarsSu0618]

@makseq Because our data sometime have different means according to everyone and need to relabel data.
if it have the search bar function , it can increase user experiences and decrease search cost.

[makseq]

@MarsSu0618 Ok, I've added it to our backlog, it's a good idea. But it's not relevant for this thread :slightly_smiling_face:

So, what about your previous problem and my previous question?

I see that you have a checkbox "Treat every bucket object as source file ...". In your case it must be off. Maybe something change its state unexpectedly?

Do you still have this problem?

[makseq]

Feel free to reopen if your problem is not gone.