Model not converging

[dat4git]

Hi,
I'm trying to run slalom ('1.8.0') on a fairly large dataset ~12K cells x 20K genes with 50 annotated factors. I used scater + scran to qc the data as recommended and generated a couple of subsets using varying hvg thresholds. When I run slalom with (largely) default settings the model fails to converge in most cases despite a large number of iterations. (listed below)

Could you please provide some insights/guidance on what parameters aid model convergence and what causes it to fail? Thanks!!

model <- newSlalomModel(sce, geneset, n_hidden = 5,
min_genes = 20, prune_genes=TRUE)
model <- initSlalom(model)
model <- trainSlalom(model, nIterations=10000) # or 5000

Failed to converge:
[1] 4169 hvgs x 11634 cells
35 annotated factors retained; 15 annotated factors dropped.
908 genes retained for analysis.
initiating model
training model
pre-training model for faster convergence
iteration 0
Model not converged after 50 iterations.
iteration 0
Model not converged after 50 iterations.
iteration 0
iteration 100
..
iteration 10000
Model not converged after 10000 iterations.

running on half the cells and half num genes
[1] 2838 hvgs x 5817 cells
30 annotated factors retained; 20 annotated factors dropped.
627 genes retained for analysis.
Model not converged after 5000 iterations.

Instance where model converged :
1735 hvgs x 11634 cells
11 annotated factors retained; 39 annotated factors dropped.
209 genes retained for analysis.
Model converged after 3200 iterations.

Note: A similar test dataset with 9057 hvgs and 6302 cell also converged