hanzhaoml / mdan Goto Github PK

Hi,

I am getting the following error when trying to run the code with the default settings in config file for MNIST dataset:
ValueError: Cannot feed value of shape (48, 32, 32, 3) for Tensor u'feature:0', which has shape '(256, 32, 32, 3)'
Can somebody please help?
Also can you confirm version of Python, PyTorch and tensorflow-gpu used?

Thanks!

Do you plan to add details on what you changed in the code to extend MDAN to the WebCamT regression experiments? I am very interested in trying this.

https://github.com/KeiraZhao/MDAN/blob/4eaec5b0d49a3af446b3b52850f646dfed507a14/main_amazon.py#L143

Would it rather be:

loss = torch.max(losses + mu * torch.min(domain_losses))

to correspond with Eq(5) of https://papers.nips.cc/paper/8075-adversarial-multiple-source-domain-adaptation.pdf?

Hello, thank you for sharing your code and works.
I'm trying to reproduce your digits dataset result, but it's hard to reproduce it.
In config file, i changed source_num to 3, source_type1,2,3 for MNIST, MNIST_M, SYNTHDIGITS and target_type1 for SVHN. I checked datasets are downloaded properly, but the accuracy is around 20%. I checked out your default config setting works well with ~98% accuracy (MNIST->MNIST). Is there any other things to be modified? Thank you.

Hi,

I have changed this project in order to try domain adversarial learning in a regression problem. actually, I replaced softmax - nll_loss with MAE. But it just makes the regression accuracy worse.
Do you have any idea why it happens?

from TensorflowToolbox.utility import file_io

ImportError: No module named 'TensorflowToolbox'

Dear Sir, how to solve the above problem?