Heatmap issue about deeptemporalclustering HOT 4 OPEN

Hi,
I assume you are displaying the heatmap dimension corresponding to the predicted cluster for each sample. Are the two samples assigned to the same or different clusters?

The heatmap is indicating which region (time steps) of the input contributed most to being assigned to each cluster, so in the first example, the entire time series contributed equally, while in the second, the central part (range 60-110) contributed most.

I suggest also to look at the cluster centroids to interpret the results.

from deeptemporalclustering.

Thanks Florent for your response.
Yes the heatmap dimension is derived using the predicted cluster id for each sample. Both samples are from the same cluster.
I was trying to extract data windows with an event/Anomalous windows using heatmap values ( like if heatmap value breaches above 90% max heatmap value in a cluster ) and loss threshold. Any suggestion on this?

I will look into the cluster centroids .

from deeptemporalclustering.

How can i generate a heatmap or heatmaps ?

from deeptemporalclustering.

@JJQKA2, see for instance this issue

from deeptemporalclustering.