Skip to content

Latest commit

 

History

History
16 lines (15 loc) · 2.95 KB

File metadata and controls

16 lines (15 loc) · 2.95 KB
Field Expected Datatype Descriptions
input-path str The path to the Dataset
output-path str Location to flush all outputs
model-path str The path to the Model
maa-model bool Whether to use the Masked Amino Acid Model, for details, seeTCR-BERT's white paper
negative-dir list[str] Directory Location to Control Data
positive-dir list[str] Directory Location to Cancer Patients
batch-size int Amount of TCR sequences inside each patient to pass into the model at once
epoch int Amount of Epochs to train the model
lr float or list[float] Learning Rate, or a List of Learning Rates
change-lr-at float or list[float] Epochs to change Learning Rate at. This should be the same datatype as "lr"
train-split float The proportion of data to be served as the training data
bag-accummulate-loss int The amount of patients to incur a step down the gradient
l2-penalty float Amount of Weight Decay