# Parse config file
if args.neptune_config is not None:
with open(args.neptune_config) as neptune_file:
neptune_config = yaml.safe_load(neptune_file)
utils.initialize_neptune(neptune_config)
all_data, labeled_data, unlabeled_data = datasets.load_binary_data(
args.dataset_config, args.label, args.negative_class)
# Load binary data subsets the data to two classes. Update the label encoder to treat this
# data as binary so the F1 score doesn't break
labeled_data.recode()
label_encoder = labeled_data.get_label_encoder()
# Correct for batch effects
if args.batch_correction_method is not None:
all_data = datasets.correct_batch_effects(all_data,
args.batch_correction_method)
labeled_data = all_data.get_labeled()
labeled_data.subset_samples_to_labels(
[args.label, args.negative_class])
unlabeled_data = all_data.get_unlabeled()
# Get fivefold cross-validation splits
labeled_splits = labeled_data.get_cv_splits(num_splits=args.num_splits,
seed=args.seed)
# Train the model on each fold
accuracies = []
balanced_accuracies = []
f1_scores = []
supervised_train_studies = []
supervised_train_sample_names = []
with open(args.dataset_config) as data_file:
dataset_config = yaml.safe_load(data_file)
with open(args.simulation_config) as data_file:
simulation_config = yaml.safe_load(data_file)
all_data, labeled_data, unlabeled_data = datasets.load_binary_data(
args.dataset_config, args.label, args.negative_class)
# Load healthy + disease data
expression_data = labeled_data.subset_samples_to_labels(
[args.label, args.negative_class])
# Correct for batch effects
if args.batch_correction_method is not None:
expression_data = datasets.correct_batch_effects(
expression_data, args.batch_correction_method)
# Scale data
labeled_scaler = preprocessing.MinMaxScaler()
train_data, _ = expression_data.get_all_data()
labeled_scaler.fit(train_data)
learning_rate = simulation_config['lr']
batch_size = simulation_config['batch_size']
epochs = simulation_config['epochs']
kappa = simulation_config['kappa']
intermediate_dim = simulation_config['intermediate_dim']
latent_dim = simulation_config['latent_dim']
epsilon_std = simulation_config['epsilon_std']
val_frac = simulation_config['val_frac']
