Shared output_dir and random_seed fields are documented in Library configuration. Model-specific fields for type: mofa:
| Key | Type | Default | Description |
|---|---|---|---|
n_factors |
int |
50 |
Number of factors to infer. Must be positive. |
ard_factors |
bool |
true |
Applies automatic relevance determination to the factors. |
drop_r2 |
float or null |
0.001 |
Drops factors that explain no more variance than this threshold in every view and group. null keeps every factor. |
scale_views |
bool |
false |
Scales views to unit variance before fitting. |
scale_groups |
bool |
false |
Scales groups to unit variance before fitting. |
convergence_mode |
str |
fast |
Convergence tolerance. Options are fast, medium, or slow. |
iterations |
int |
1000 |
Maximum number of training updates. Must be positive. train warns when a run stops at this cap without converging. |
gpu_mode |
bool |
false |
Enables GPU training through CuPy. Requires an NVIDIA GPU and a CuPy installation compatible with the server's CUDA version. mofapy2 falls back to CPU if CuPy cannot be imported. |
gpu_device |
int or null |
null |
CUDA device index (non-negative), used when gpu_mode: true. null uses CuPy's current device. |