##rna_X: RNA-seq data; dna_X: ATAC-seq data
##Both rna_X and dna_X are matrices in which each row represents one cell while each column stands for a gene
##rna_X and dna_X don't have to have the same number of cells
##rna_X as input for source_trainset while dna_X as input for target_trainset

##Integration via sciCAN
import torch
import torch.nn.functional as F
from sciCAN.network import *
from sciCAN.training import *

FeatureExtractor = Cycle_train_wolabel(epoch=100, batch_size=1024, source_trainset=rna_X, target_trainset=dna_X)

##Visualization integration outcome
X_tensor_a = torch.tensor(rna_X).float()
X_tensor_b = torch.tensor(dna_X).float()
FeatureExtractor.to(torch.device("cpu"))
X_all_tensor = torch.cat((X_tensor_a,X_tensor_b),0)
y_pred = FeatureExtractor(X_all_tensor)
y_pred = F.normalize(y_pred, dim=1,p=2)
y_pred = torch.Tensor.cpu(y_pred).detach().numpy()##visualize y_pred with UMAP