adversarial-autoencoder
adversarial-autoencoder copied to clipboard
musyoku
dependency environment problem
Hello! I use anaconda spyder to compile your perfect project, but I meet some problems. I think it's about my compile environment. I use win7 64bit, and Chainer version is 2.01, cupy version 4.1, cuda 9.0, cudnn 7. There are always some problems about the carray.cuh of cupy. Could you please tell me your compile environment and do u know what about this problem?
I am using Ubuntu 16.04 and CUDA 9.0.
I installed cupy via pip install cupy-cuda90
Could you show me the error message?
Thanks for your kindly reply in time. I tried it in the linux system. Luckily, everything is OK. Thank u again!
Hello musyoku! I 'm a new hand. I have learned your perfect project, and handle the example successfully. And here, several problems puzzles me always. Could you please give some hints? Here,
(1) I work on the unsupervised code, and it runs smoothly. Here, I don't known what the arguments 'ndim_y=16, ndim_z=10, ndim_h=1000' (in the model.py) refer to? The ndim_z refers to number of cluster? right? What about others?
(2)When I get the unsupervised model, then I can visualize to get the cluster.png. The first column is generated date? (not from the original dataset?) And the other chips is real data from the dataset? they are clustered according to the first column?
(1) ndim_y is the number of clusters, ndim_z is the style dimension, ndim_h is hidden layer dimension.
(2) Each row corresponds to one cluster. The first column is the cluster head and the other images are from dataset.
Thank you very much ! The head of each row is decoded from the latent style vector, right? And they are behalf of the maximum difference style of the setting dimy, can i understand in this way?
(1) No. The cluster head is decoded from the one-hot vector y.
The first image in each row shows the cluster heads, which are digits generated by fixing the style variable to zero and setting the label variable to one of the 16 one-hot vectors.
(2) I think that it is a class difference rather than a style difference because style vector is fixed to zero.