Thank you very much for your work. When I am clustering image data on a custom dataset, I am using a pre-trained AE for feature downscaling from 256 dimensions to 20 dimensions and then alternating between feature extraction and clustering operations. However all the samples are categorized into one cluster, even though I made the following changes during the clustering process, the final clustering result is that all the samples are still categorized into one cluster, is there any other way to make the splitting more adequate?
The programs you have previously offered to promote splitting are：
- increase alpha (e.g., alpha=1000, or 10,000)
and/or
- Pick Psi to be small (e.g., take the default Psi and scale it by a small number) AND set nu to be very high (using a small Psi without a large nu will not help). The reason is that this will mean you are condifent that you want samll clusters, so the model will be driven to use many of them in order to fully cover all the data.
Thank you very much for your work. When I am clustering image data on a custom dataset, I am using a pre-trained AE for feature downscaling from 256 dimensions to 20 dimensions and then alternating between feature extraction and clustering operations. However all the samples are categorized into one cluster, even though I made the following changes during the clustering process, the final clustering result is that all the samples are still categorized into one cluster, is there any other way to make the splitting more adequate?
The programs you have previously offered to promote splitting are：
and/or