scDeepCluster
scDeepCluster copied to clipboard
ttgump
difference
What is the difference between scDeepCluster.py,scDeepCluster_latent.py and scDeepCluster_estimate_k.py?
What is the difference between scDeepCluster.py,scDeepCluster_latent.py and scDeepCluster_estimate_k.py?
"scDeepCluster.py" is the code of scDeepCluster model. "scDeepCluster_latent.py" is the code used to extract latent embedding after the model training. "scDeepCluster_estimate_k.py" is the code to use an intuitive metric generalizability to estimate the number of clusters from data.
Thank you very much for your reply to my question, but I still have a doubt. The network structure diagram run out with the code you gave seems to be different from the network structure diagram you gave on github. The main difference is in the way of adding Gaussian noise. Attached below is the diagram you gave with a diagram of the code running.
---- Replied Message ---- | From | Tian @.> | | Date | 03/24/2024 02:37 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [ttgump/scDeepCluster] difference (Issue #10) |
What is the difference between scDeepCluster.py,scDeepCluster_latent.py and scDeepCluster_estimate_k.py?
"scDeepCluster.py" is the code of scDeepCluster model. "scDeepCluster_latent.py" is the code used to extract latent embedding after the model training. "scDeepCluster_estimate_k.py" is the code to use an intuitive metric generalizability to estimate the number of clusters from data.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
Sorry, I sent it again, and I found that you may not be able to open the image link on github, you can try to view it directly in the mailbox. Thank you!
---- Replied Message ---- | From | Tian @.> | | Date | 03/26/2024 01:01 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [ttgump/scDeepCluster] difference (Issue #10) |
Sorry I can't see your attachment, could you send it again? Thanks.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
It seems I didn’t receive your attachment. Could you send it to another email box: @.@.>. Thanks. Best, Tian
From: GGyynnn @.> Date: Tuesday, March 26, 2024 at 9:50 AM To: ttgump/scDeepCluster @.> Cc: Tian Tian @.>, Comment @.> Subject: Re: [ttgump/scDeepCluster] difference (Issue #10) Sorry, I sent it again, and I found that you may not be able to open the image link on github, you can try to view it directly in the mailbox. Thank you!
---- Replied Message ---- | From | Tian @.> | | Date | 03/26/2024 01:01 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [ttgump/scDeepCluster] difference (Issue #10) |
Sorry I can't see your attachment, could you send it again? Thanks.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
— Reply to this email directly, view it on GitHubhttps://github.com/ttgump/scDeepCluster/issues/10#issuecomment-2019241776, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AAJCBCYD37JRWKO4RLJ6AUTY2DH6JAVCNFSM6AAAAABFDE6YHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMJZGI2DCNZXGY. You are receiving this because you commented.Message ID: @.***>
I'm sorry. I know it was a little abrupt. I found that github can upload pictures. The following are the network structure you gave on github and the network structure I ran out with your code. The main difference seems to be the way Gaussian noise is added. Does that make a difference? I am looking forward to your reply.
Yes, there is difference. The main difference is I didn't add noise to the clustering layer, which will make the model difficult to converge. Why don't you use the PyTorch version? This code is based on Keras 1.0 which is not supported any more currently.
I'm sorry, I don't understand what you mean. You said that you did not add noise to the clustering layer, but because I modified the network to remove the clustering layer. The way I'm talking about noise being added differently is during encoding . As shown in the following figure. Also thanks for your advice, I will try the PyTorch version. Can you answer the questions I have raised about this code? What causes these two different ways of adding noise? Is it the structure of the network or the keras version? Do different noise adding methods have a big impact on the overall network?