search

stefanradev93 / DataSizeMatters

Contains the scripts, pre-trained networks, and data for our paper:
19 stars 4 forks source link

order to run the files and tensorflow version #1

Open ljr10225008 opened 1 year ago

ljr10225008 commented 1 year ago

Hi

Next to meet you. I am also studying the response time of humans in certain tasks and found that your model is very interesting and will be helpful to analyze our data. Thus, we are expecting to refer to your model. However, when we try to run the codes that are shared in the paper, several problems come out.

If it is convenient, could you please tell us how to solve the following problems?

  1. What is the order to run the scripts in the folder? There are "IAT_BayesFlow_TrainingPhase" and "IAT_BayesFlow_InferencePhase" files. Should I run the "training phase" first, and then run " inference phase", or I can directly use the "inference phase" to analyze the data? We need to modify some parameters in the models.
  2. There are files named like "...__Collapsing". What are they used for?
  3. When I run the training phase, it came out that the problems "name 'global_step' is not defined" and an Optimizer version problem. I guess it is because of the wrong version of TensorFlow we used. Thus, could you tell us the version of python and the packages, e.g., Tensorflow, sklearn, etc. you used for these codes?
  4. When running the inference phase, it came out that the file "outcomes_with_id_161718.csv" can not be found, but we did not find a file named like this in the shared osf dataset. Could you tell me which data I can use in your shared folder to test the codes？

Below are the screenshots of the errors image image Thank you!

Jiarui

stefanradev93 commented 1 year ago

Hi Jiarui,

The way I see it, you need to re-train the networks for inference based on a new model. This project used quite an older TensorFlow version (tensorflow 1.13). Many things are different now, so if you want to train new networks for inference, I strongly suggest you look at our new library:

https://github.com/stefanradev93/BayesFlow/tree/Development

And maybe try out some of the linked walk-throughs. We will be porting this code into the new library, but this may take some time. Happy to assist further.

Cheers, Stefan

ljr10225008 commented 1 year ago

Many thanks! I tried the bayesflow library and it works super well! We'll try to get some ideas from the walk-thoughts and train our model.

Jiarui

stefanradev93 commented 1 year ago

Glad to hear that, Jiarui! Let me know how that goes and if you have any feedback. It's an open-ended project. :)