'PJFDataset' object has no attribute 'geek_direct' Error

[benm-dev]

Hello when i run DPGNN with the kaggle jobrec dataset i get this error

i am using recbole 1.0.0 and all the requirements are correct

[flust]

Could you please show me your configuration file? (overall.yaml and jobrec.yaml)

[benm-dev]

okay so the error has changed to this now i upgraded my recbole version

Here are the config files

[flust]

I fixed the problem, you can pull the latest code to try again.

This problem is because in the latest version of RecBole, there is a default configuration file "sample.yaml" for all dataset and in this file, the "DIRECT_FIELD" is defined as "field".

We're also doing some code updates and maintenance recently and If you have any more questions, please feel free to contact me~

[benm-dev]

Hey so i am not receiving the direct error anymore but I am now receiving this error again

Here are the config files again

[flust]

This means that the model DPGNN requires active behavior on both sides in the data, and therefore, it requires the data to have the feature "DIRECT_FIELD".(Refer to the original paper, it need directed behavior to construct the interaction graph)
In order to achieve this, in the "DIRECT_FIELD" feature, we use '0' to represent the user's active behavior("geek_direct" in the code), and "1" to represent the employer's active behavior, which may not available in the jobrec dataset.

[flust]

We find that there are still some bugs for model DPGNN when we test it. We will fix it in few days.

…

---Original--- From: "Ben ***@***.***> Date: Mon, Nov 14, 2022 06:46 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [RUCAIBox/RecBole-PJF] 'PJFDataset' object has no attribute'geek_direct' Error (Issue #3) Hey so i am not receiving the direct error anymore but i am now receiving this error again Here are the config files again — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[benm-dev]

okay thanks so much so there aren't any other English datasets I could use with the DPGNN Model

[flust]

I'm afraid so now. We will try to solve this problem as soon as possible!

…

---Original--- From: "Ben ***@***.***> Date: Wed, Nov 16, 2022 07:09 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [RUCAIBox/RecBole-PJF] 'PJFDataset' object has no attribute'geek_direct' Error (Issue #3) okay thanks so much so there aren't any other English datasets I could use with the DPGNN Model — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[benm-dev]

hey so i am now trying to run DPGNN with the zillhan dataset and i am still receiving the same error but zillhan does have the feature direct filed

[benm-dev]

i also receive this error when i run APJFNN with zhilian

do you guys have any other datasets that could be used with this
like the boss_search dataset

[flust]

Sorry it took so long to reply.

We check the 'yaml' file and fix bugs in the code.
We updated the code and provided configuration files that can be run directly with dataset zhilian.

For the model DPGNN, the experimental datasets we used are not available for the privacy policy of the company.
In order for you to run the model code properly, we have modified the zhilian data processing procedure.
Before that, it takes a long time for computing bert encoding before. So in order to run at an acceptable time, we just keep a small part of it. If you don't need the bert part, just set it to false in the 'DPGNN.yaml'.
In addition, We randomly added directional information to this dataset to run DPGNN correctly.
You can now run the DPGNN model directly under the right environment.

Besides, this dataset is just an example and the features it contains may not necessarily fit all models.
For better experiments, we recommend that you work with your own dataset and the format can be referred to zhilian.

Moreover, most bugs are due to configuration files not being set up properly(including 'overall.yaml', 'dataset/zhilian.yaml' and 'model/DPGNN.yaml') like the bug of "direct_field". You can check them first if you run into any other problems.

Sorry again for taking so long to reply.