when run the code:
class ClusteringLayer(Layer):

def __init__(self, n_clusters, weights=None, alpha=1.0, **kwargs):
    if 'input_shape' not in kwargs and 'input_dim' in kwargs:
        kwargs['input_shape'] = (kwargs.pop('input_dim'),)
    super(ClusteringLayer, self).__init__(**kwargs)
    self.n_clusters = n_clusters
    self.alpha = alpha
    self.initial_weights = weights
    self.input_spec = InputSpec(ndim=2)

def build(self, input_shape):
    assert len(input_shape) == 2
    input_dim = input_shape[1]
    self.input_spec = InputSpec(dtype=K.floatx(), shape=(None, input_dim))
    self.clusters = self.add_weight((self.n_clusters, input_dim), initializer='glorot_uniform', name='clusters')
    if self.initial_weights is not None:
        self.set_weights(self.initial_weights)
        del self.initial_weights
    self.built = True

def call(self, inputs, **kwargs):
    q = 1.0 / (1.0 + (K.sum(K.square(K.expand_dims(inputs, axis=1) - self.clusters), axis=2) / self.alpha))
    q **= (self.alpha + 1.0) / 2.0
    q = K.transpose(K.transpose(q) / K.sum(q, axis=1)) # Make sure each sample's 10 values add up to 1.
    return q

def compute_output_shape(self, input_shape):
    assert input_shape and len(input_shape) == 2
    return input_shape[0], self.n_clusters

def get_config(self):
    config = {'n_clusters': self.n_clusters}
    base_config = super(ClusteringLayer, self).get_config()
    return dict(list(base_config.items()) + list(config.items()))

clustering_layer = ClusteringLayer(n_clusters, name='clustering')(encoder.output)
model = Model(inputs=encoder.input, outputs=clustering_layer)

TypeError: add_weight() got multiple values for argument 'name'

have you tried using this on a tabular data, e.g. iris ? (with a smaller number of nodes and parameters). do you think, makes sense to try it out ?

Hello, I'm not an expert in this field however I'm trying to do something on it. One simple question: Why are you using x as x=np.concatenate (x_train,x_test)? How can you check the NN's performance if you don't separate x_train from x_test?
Thanks in advance

Thank you for sharing the code.
I am new in Keras. I know it is a stupid idea to ask this bellow question.
In ClusteringLayer, the encoder.output shpae is (?,10) which has been passed through the custom layer. When I pass the encoder.output weights to the custom layer "Bulid function", it generates weights for self.clustering by the size of(10,10) i.e the n_clusters is 10 and input_shape[1] is 10. Is it so? or I am missing something.
In "Call function", specifically in this line, q = 1.0 / (1.0 + (K.sum(K.square(K.expand_dims(inputs, axis=1) - self.clusters), axis=2) / self.alpha)), why do you use K.expand_dims? What is the purpose of using this?

Please help me to figure out these two problems.
Again thank you for sharing the beautiful code.

Hi!
Very interested in using your code, but you have not attached a license!

Could you perhaps do so? The BSD and MIT licenses are probably the most used in the scientific python community.

Thanks!

rt

when I try to run the part
Model to train clustering and autoencoder at same time(Fully connected)!
I get the
ValueError: Layer AE was called with an input that isn't a symbolic tensor. Received type: <class 'int'>. Full input: [784, 500, 500, 2000, 10]. All inputs to the layer should be tensors.
pls hel fix it

What if we don't know the number of clusters ?! and if the case is something like tabular data (many users with demographical data) do we need the autoencoder model ?!
How I can invest your code in data as I mentioned?

Hey!
First off, thank you so much for sharing your code- it's super helpful :)

I wanted to know why the loss is 0 for all the iterations, during the evaluation stage, and if there is any implication regarding the same.

Hello,
How do you save the results of the model into their respective cluster folders? In this case there is 10 clusters so then there should be 10 folders with images from the clusters inside.

There is a data set that only knows the category but does not have an annotation for each image. How to evaluate it?

Thanks your excellence work! It helps me a lot, and I wonder why the why the results far overtakes the state of DEC paper, 96.2% VS 84.3%? Is there any tricks in the experiment？