Generating Spectra factors for test set

[mariemoullet]

hello, thank you very much for this method and tutorials!

i was wondering if there was the option to train the spectra model on one dataset and then generate spectra factors / scores for a new dataset?

following the syntax of the tutorial, something like

model = spc.SPECTRA_Model(X = X_train, labels =  np.array(adata_train.obs['cell_type']), 
                          L = L,
                          adj_matrix = adj_matrix,
                          weights = None, lam = 0.01,  delta=0.001,kappa = None, rho = 0.05, 
                          use_cell_types = True)


model.train(X = X_train, labels =  np.array(adata_train.obs['cell_type']),
            num_epochs=2) 

model.test(X = X_test, labels =  np.array(adata_test.obs['cell_type'])) 

thank you very much in advance!

[wallet-maker]

Great question, this is not implemented at the moment.

If the genes are identical in training in test it should be possible to do this by taking the model object output and setting requires_grad=False in the training.

Let me know if this works. Also, I think it would be really helpful to find a solution here which also works when the gene space is not identical in training and test.