Skip to main content
. Author manuscript; available in PMC: 2019 Mar 14.
Published in final edited form as: Pac Symp Biocomput. 2019;24:362–373.

Table 1.

Best and worst parameter values for two- and three- layer Tybalt models with many cells for simulated datasets. l: learning rate, b: batch size, e: epoch, c: dimensionality of the first hidden layer.

2-layer model
Best combination Worst combination
ncells l b e c NMI ARI l b e c NMI ARI
10000 0.0005 200 25 100 0.99 0.99 0.0005 50 200 500 0.08 0.05
20000 0.001 200 25 250 0.98 0.95 0.001 200 200 500 0.2 0.13
30000 0.0005 100 100 100 0.97 0.97 0.0005 50 100 500 0 0
3-layer model
Best combination Worst combination
ncells l b e c NMI ARI l b e c NMI ARI
10000 0.001 50 100 100 1 1 0.002 200 200 500 0.06 0.04
20000 0.0005 100 25 250 0.97 0.95 0.001 100 100 500 0.02 0.01
30000 0.0005 100 25 250 0.96 0.94 0.0005 50 200 250 0 0