if args.select_on is None and args.select_method is None:
result_dir = result_prefix + os.sep + "no_feature"
else:
result_dir = result_prefix+os.sep+args.select_method+'_'+\
str(args.n_features)+'_on_'+args.select_on
os.makedirs(result_dir, exist_ok=True)
load_ind, train_adata, test_adata = load_adata(result_dir)
if not load_ind:
train_adata, test_adata = dataloading_utils.load_Pancreas_adata(
data_dir, result_dir, args=args)
## whether to purify reference dataset
purify_method = ""
if "purify_dist" in args.data_source:
purify_method = "distance"
elif "purify_SVM" in args.data_source:
purify_method = "SVM"
train_adata, test_adata = dataloading_utils.process_loaded_data(
train_adata,
test_adata,
result_dir,
args=args,
purify_method=purify_method)
print("Train anndata: \n", train_adata)
print("Test anndata: \n", test_adata)
method_utils.run_pipeline(args, train_adata, test_adata, data_dir,
result_dir)
print("Test anndata: \n", test_adata)
if args.downsample: ## add shuffled cells
train_cells = train_adata.obs_names.tolist()
random.seed(args.sample_seed)
random.shuffle(train_cells) ## shuffle original cell list
original_train_adata = train_adata.copy()[train_cells]
for i in range(original_train_adata.shape[0] // args.downsample_size +
1):
sampled_number = (i + 1) * args.downsample_size if (
i + 1) * args.downsample_size < original_train_adata.shape[
0] else original_train_adata.shape[0]
train_adata = original_train_adata[:sampled_number]
sampled_result_dir = result_dir + os.sep + str(sampled_number)
os.makedirs(sampled_result_dir, exist_ok=True)
method_utils.run_pipeline(args, train_adata, test_adata, data_dir,
sampled_result_dir)
else: ## add shuffled individuals
if args.data_source == "mousebrain_crossdataset_inds": ## a combined version
pFC_samples = [
x for x in train_adata.obs["Sample"].tolist() if x != 'nan'
]
allen_samples = [
x
for x in train_adata.obs["external_donor_name_label"].tolist()
if x != 'nan'
]
train_adata.obs["ind"] = pFC_samples + allen_samples
original_train_adata = train_adata.copy()
train_inds = list(set(original_train_adata.obs['ind']))
random.seed(args.sample_seed)