search

Philyzh8 / scTAG

"ZINB-based Graph Embedding Autoencoder for Single-cell RNA-seq Interpretations" in AAAI 2022
MIT License
20 stars 2 forks source link

Gradients do not exist for variables ['clustering/clusters:0'] when minimizing the loss. #1

Open zzdj11 opened 2 years ago

zzdj11 commented 2 years ago

I tried to change to another dataset, but tensorflow came up with the following warning during training and couldn't find a solution. This bothers me a lot. WARNING:tensorflow:Gradients do not exist for variables ['clustering/clusters:0'] when minimizing the loss.

Philyzh8 commented 2 years ago

Are you using single-cell data? scTAG is more suitable for single-cell data with strong sparsity. If not, you can also do some feature selection processing on the data.

zzdj11 commented 2 years ago

Are you using single-cell data? scTAG is more suitable for single-cell data with strong sparsity. If not, you can also do some feature selection processing on the data. Thanks for your answer! I am using spatial resolved transcriptomics data, which also has strong sparsity. What do you suggest I should check to resolve this warning? The warning code appears as the gradient update statement: optimizer.apply_gradients(zip(grads, vars))

tb1over commented 2 years ago

i think you should compare your model performance with scDeepCluster, which you cited. your work add a graph autoencoder module baes the scDeepCluster