@XifengGuo, nice code! Thank you for sharing it. Would you add a licence, please?

论文中说了p的计算是依赖于q，按照我自己的理解，p应该是真实分布，为什么用q去计算可行呢？以及公式8对p的计算方法的依据是什么？期待回答

Hello, I am very interested in your article, but what is the reason why the accuracy rate of code running on my computer is always around ten percent

I tried running your code on mnist dataset itslef without any chnage in code but received following error -

self.clusters = self.add_weight((self.n_clusters, input_dim), initializer='glorot_uniform', name='clusters')
TypeError: add_weight() got multiple values for argument 'name'

Please help.

I read your paper works.
I seems that this work was so interesting.
I'm planning to apply in the field of natural language processing.

However, I'm not sure to how to visualize clustering results like yours .

你好，想问下您从哪里可以获取到聚类的结果！

Hi, Nice work!
I would like to know if you tried and get some results on dataset STL10.
Thanks

I have the next message when I run

bash ./download_usps.sh

--2019-11-01 17:07:48-- http://www-i6.informatik.rwth-aachen.de/~keysers/usps_train.jf.gz
Resolvendo www-i6.informatik.rwth-aachen.de (www-i6.informatik.rwth-aachen.de)... 137.226.36.140
Conectando-se a www-i6.informatik.rwth-aachen.de (www-i6.informatik.rwth-aachen.de)|137.226.36.140|:80... conectado.
A requisição HTTP foi enviada, aguardando resposta... 404 Not Found
2019-11-01 17:07:49 ERRO 404: Not Found.

gunzip: can't stat: usps_train.jf.gz (usps_train.jf.gz.gz): No such file or directory
--2019-11-01 17:07:49-- http://www-i6.informatik.rwth-aachen.de/~keysers/usps_test.jf.gz
Resolvendo www-i6.informatik.rwth-aachen.de (www-i6.informatik.rwth-aachen.de)... 137.226.36.140
Conectando-se a www-i6.informatik.rwth-aachen.de (www-i6.informatik.rwth-aachen.de)|137.226.36.140|:80... conectado.
A requisição HTTP foi enviada, aguardando resposta... 404 Not Found
2019-11-01 17:07:50 ERRO 404: Not Found.

gunzip: can't stat: usps_test.jf.gz (usps_test.jf.gz.gz): No such file or directory

When looking at the counts in each cluster after each update interval, iterations keep continuing until one giant cluster remains or there a few clusters but empty clusters.
Does this always happen? Is the key stopping the algorithm at a good point before this happens?
( I am working with 2D image data. )
Thank you.

is there a way to run your code on GPU? because my data set is over 1000000 images and takes forever to run on CPU

I try to replicate this code on Conv3D and I get error about the input dimension. I already modified the input shape for the 3D ConvNet and it seems this has to do with the clustering hidden layer. I am at a loss on where to rework the code to get it running.

`Layer (type) Output Shape Param #

conv3d_19 (Conv3D) (None, 55, 55, 10, 128) 15616

conv3d_20 (Conv3D) (None, 26, 26, 10, 64) 204864

embedding (ConvLSTM2D) (None, 26, 26, 10, 64) 295168

conv_lst_m2d_9 (ConvLSTM2D) (None, 26, 26, 10, 32) 110720

conv_lst_m2d_10 (ConvLSTM2D) (None, 26, 26, 10, 64) 221440

conv3d_transpose_9 (Conv3DTr (None, 55, 55, 10, 128) 204928

conv3d_transpose_10 (Conv3DT (None, 227, 227, 10, 1) 15489

Total params: 1,068,225
Trainable params: 1,068,225
Non-trainable params: 0

ValueError Traceback (most recent call last)
in ()
81
82 # prepare the DCEC model
---> 83 dcec = DCEC(input_shape=(227,227,10,1), n_clusters=n_clusters)
84 plot_model(dcec.model, to_file= save_dir + '/dcec_model.png', show_shapes=True)
85 dcec.model.summary()
in init(self, input_shape, n_clusters, alpha)
80
81 # Define DCEC model
---> 82 clustering_layer = ClusteringLayer(self.n_clusters, name='clustering')(hidden)
83 self.model = Model(inputs=self.cae.input,
84 outputs=[clustering_layer, self.cae.output])
~\Anaconda2\envs\py36\lib\site-packages\keras\engine\base_layer.py in call(self, inputs, **kwargs)
441 # Raise exceptions in case the input is not compatible
442 # with the input_spec set at build time.
--> 443 self.assert_input_compatibility(inputs)
444
445 # Handle mask propagation.
~\Anaconda2\envs\py36\lib\site-packages\keras\engine\base_layer.py in assert_input_compatibility(self, inputs)
309 self.name + ': expected ndim=' +
310 str(spec.ndim) + ', found ndim=' +
--> 311 str(K.ndim(x)))
312 if spec.max_ndim is not None:
313 ndim = K.ndim(x)
ValueError: Input 0 is incompatible with layer clustering: expected ndim=2, found ndim=5`