Model Card for maps_bis
This model was trained with ClinicaDL. You can find here all the information.
General information
This model was trained for classification and the architecture chosen is Conv4_FC3.
Model
architecture: Conv4_FC3
multi_network: False
ssda_network: False
Architecture
dropout: 0.0
latent_space_size: 2
feature_size: 1024
n_conv: 4
io_layer_channels: 8
recons_weight: 1
kl_weight: 1
normalization: batch
Classification
selection_metrics: ['loss']
label: diagnosis
label_code: {'AD': 0, 'CN': 1}
selection_threshold: 0.0
loss: None
Computational
gpu: True
n_proc: 32
batch_size: 32
evaluation_steps: 20
fully_sharded_data_parallel: False
amp: False
Reproducibility
seed: 0
deterministic: False
compensation: memory
track_exp:
Transfer_learning
transfer_path: ../../autoencoders/exp3/maps
transfer_selection_metric: loss
nb_unfrozen_layer: 0
Mode
use_extracted_features: False
Data
multi_cohort: False
diagnoses: ['AD', 'CN']
baseline: True
normalize: True
data_augmentation: False
sampler: random
size_reduction: False
size_reduction_factor: 2
caps_target:
tsv_target_lab:
tsv_target_unlab:
preprocessing_dict_target:
Cross_validation
n_splits: 5
split: []
Optimization
optimizer: Adam
epochs: 200
learning_rate: 1e-05
adaptive_learning_rate: False
weight_decay: 0.0001
patience: 10
tolerance: 0.0
accumulation_steps: 1
profiler: False
save_all_models: False
Informations
emissions_calculator: False
Other information
latent_space_dimension: 64
preprocessing_dict: {'preprocessing': 't1-linear', 'mode': 'roi', 'use_uncropped_image': False, 'roi_list': ['leftHippocampusBox', 'rightHippocampusBox'], 'uncropped_roi': False, 'prepare_dl': False, 'file_type': {'pattern': '*space-MNI152NLin2009cSym_desc-Crop_res-1x1x1_T1w.nii.gz', 'description': 'T1W Image registered using t1-linear and cropped (matrix size 169×208×179, 1 mm isotropic voxels)', 'needed_pipeline': 't1-linear'}}
mode: roi
network_task: classification
caps_directory: $WORK/../commun/datasets/adni/caps/caps_v2021
tsv_path: $WORK/Aramis_tools/ClinicaDL_tools/experiments_ADDL/data/ADNI/train
validation: KFoldSplit
num_networks: 2
output_size: 2
input_size: [1, 50, 50, 50]