Extra Information!!!

[WilliamPham1602]

These are some topics, which required your help:

• Application information ( Web or install an application for Mac or Windows). Note: because I don't have any experience in application development so I am quite worried about it.
• My plan (next step) is to use one model with 2 inputs (the first input is used vgg16 to extract features and the second is used lstm), because my base model worked so this next step is to apply extra computer vision (vgg16). Do I need-aware any detail on this step?

Thank you for your help.

[JoaoLMPereira]

Hi @WilliamPham1602 ,

1. What do you expect my help to be with application information? Do you mean how do you develop applications? My own setup is a linux distribution installed with eclipse (I started as a java dev) and plugin for Python, but there are a lot of other solutions, some more popular like pycharm.

2. I am not an expert in computer vision but as far as I known vgg16 is used for image classification, are those classes useful for this problem? Maybe there are different ways to work with it. For LSTM I don't understand if it is to apply to text or image too. If you can get some representation for a page with either of those techniques than you can try it as additional features.

[WilliamPham1602]

Hi @JoaoLMPereira,

Thanks for your quick reply!

1. Do I need to submit the application or I just demo it? What is the application input (concat image like the train set)? The description mention something like a sandbox to run an application or install it, where can I find it or which application is it?

2. For the vgg16, I will try to input the image and extract the document structure. The input of LTSM will be the content information (text). After that, I will combine two models and test them. (My ideas).

• The current solution TFIDF

Extra: Do you know where i can submit my model to test the score (like the leaderboard in Kaggle challenge).

Thanks a lots for your help.

[JoaoLMPereira]

Hi @WilliamPham1602,

Sorry about the delay.

1. I understand your question, we will provide additional information to those questions and how to submit your results shortly to everyone, please just wait and concentrate your efforts on your solution for now.

2. I understand, yes you can combine multiple techniques, specially if they represent different information. I am sure you are following recent related work in this task, so also mention that and make sure fill a research gap.

Good luck

João

[WilliamPham1602]

Hi @JoaoLMPereira,

I am wondering about the non-learned baseline model. As Maarten mention it is the mean baseline model, which compute the mean doc length, say X, and then split up the stream in X-length documents. Can you give me more detail or explanation about it?

Regards, Sang.

[JoaoLMPereira]

Hi @WilliamPham1602,

Thank you for your question. I would happily give you more details, but I think your description already says it all. Which part don't you understand?

Also, have you looked at the fixedpage function of the Evaluation.ipynb notebook that was shared?

Best,

João