scanpy.external.pp.dca results still need normolize,scale,and log1p ?

[rdf1993]

What kind of feature would you like to request?

Additional function parameters / changed functionality / changed defaults?

Please describe your wishes

If I run scanpy.external.pp.dca with default parameters, do I still need to perform normalization, scaling, and log1p afterward? How can I obtain the non-normalized values?

scanpy.external.pp.dca(adata, mode='denoise', ae_type='nb-conddisp', normalize_per_cell=True, scale=True, log1p=True, hidden_size=(64, 32, 64), hidden_dropout=0.0, batchnorm=True, activation='relu', init='glorot_uniform', network_kwds=mappingproxy({}), epochs=300, reduce_lr=10, early_stop=15, batch_size=32, optimizer='RMSprop', random_state=0, threads=None, learning_rate=None, verbose=False, training_kwds=mappingproxy({}), return_model=False, return_info=False, copy=False)

[eroell]

Hi, thanks for your interest in scanpy!

For user questions, it would be great if you could ask your question on Discourse; this is the designated discussion forum for user questions regarding scverse tools (such as scanpy).

This way, here at GitHub the focus can be put on development, while on Discourse user questions can be answered in more detail and in a manner that future users can better find previous questions.

If you think your question is related to a development issue or I misinterpreted it as a user question, we're happy to look into it here on GitHub!

Hope this gives you a guidance for receiving helpful support for your question!