File size: 1,711 Bytes
d6c7577 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 | # InterScale -- 10x Xenium human ovarian adenocarcinoma, held-out-split model.
# Full resolved config the weights in model.pt were trained with (interscale main c98a307).
# Only dataset.h5ad_data / model.save were blanked. Load with
# interscale.config.load_config('config.yaml')
dataset:
batch_size: 2
description: ''
h5ad_data: ''
layer_key: log1p_norm
mask_percentage: 0.3
mask_strategy: node
name: ovarian_xenium
num_classes: -1
num_features: -1
prediction_level: node
prediction_obs: null
prediction_task: regression
sample_key:
- sliding_window
segmentation_robustness: null
spatial_neigbors_kwargs:
coord_type: generic
library_key: sliding_window
n_neighs: 6
radius: 30
split_key: split
test_size: 0.1
train_size: 0.7
val_size: 0.2
model:
decoder:
dropout_decoder: 0.1
dual_decoder: true
hidden_dims:
- 256
- 128
type: linear
global_component:
latent_obsm_key: null
name: self-attn-transformer
parameters:
activation_func: relu
dim_feedforward: 512
dropout_global: 0.0
long_range_attention: false
max_seq_len: 3167
n_heads: 4
num_layers: 2
type_gex_embedding: null
local_component:
name: GCN
parameters:
dropout_local: 0.1
hidden_dim: 512
num_layers: 2
loss: null
n_embed: 256
save: ''
optim:
accelerator: auto
cross_corr: cell
early_stopping: true
loss: SmoothL1
lr: 0.003
lr_max_epochs: 100
lr_scheduler: CosineWarmupScheduler
lr_warmup: 10
min_delta: 0.0
min_epochs: 40
monitor: auto
n_epochs: 600
patience: 20
seed: 44
wd: 0.001
wandb:
project_name: InterScale_OvarianXenium
use: false
|