variable time step

[jzyee]

Hi in the documentation for the DTC object, found in DeepTemporalClustering.py, it is indicated that the timesteps param can be variable. However when I instantiate as follows:

dtc = DTC(n_clusters=3,
input_dim=X_train.shape[-1],
timesteps=None,
n_filters=50,
kernel_size=10,
strides=1,
pool_size=None,
n_units=[50, 1],
alpha=1,
dist_metric='eucl',
cluster_init='kmeans',
heatmap=False)

I get an error. There is an assert which brings up a typeError.

TypeError: unsupported operand type(s) for %: 'NoneType' and 'int'

Should I be using 0 instead of None?

[FlorentF9]

I assume the error is thrown by the pooling layer. You can use variable series length by setting timesteps=None (because the architecture is fully convolutional and recurrent), however you can't set pool_size=None.

[jzyee]

The following lines in DeepTemporalClustering.py throw an error:

Besides lines 55:

assert(timesteps % pool_size == 0)

Line 64:

self.latent_shape = (self.timesteps // self.pool_size, self.n_units[1])

They don't work when I assign timesteps=None

Could you show an example of how you would initiate the DTC object for a variable-length dataset

[FlorentF9]

You are right, I have to take a look at this when I have time. I keep this issue open for now.

[FlorentF9]

I took a quick look. Actually I was stuck with the variable timestep because of two things:

• Decoder: it needs to know the dimension of its input and I found no way around this. If you put a None there, then how is it supposed to know the length of the sequence to reconstruct?
• TSClusteringLayer: to initialize the cluster center weights, the dimension is needed and it depends on the timesteps. I don't see how we can define cluster centers without fixing their length with this architecture.

In a nutshell, I think the timesteps cannot be variable in DTC. But if anyone has an idea about it, please tell!