[DOCS] Update correlation decoder section & add live running example

[adelavega]

@JulioAPeraza pointed out this section may be out of date:

https://nimare.readthedocs.io/en/stable/decoding.html#correlation-based-decoding

since target_image recent changed

It would also be helpful to add a live running example with Neurosynth topics and MKDAChi2.

[JulioAPeraza]

@JulioAPeraza pointed out this section may be out of date: https://nimare.readthedocs.io/en/stable/decoding.html#correlation-based-decoding

I think this one is fine. I was looking at an outdated version of the documentation. Sorry for the confusion.

I trained a decoder locally with 50 topics for the example, but it still takes 30 minutes. I opened a PR to see if the documentation gets built with such a running time.

[adelavega]

How many iterations did you use btw?

That at least is feasible-- how many cores?

[adelavega]

Should we close this issue then?

[JulioAPeraza]

How many iterations did you use btw?

I think we don't correct for multiple comparisons the meta-analytic maps in CorrelationDecoder.

That at least is feasible-- how many cores?

I'm using 1 core. I'm not sure if we can use more in RTD.

Should we close this issue then?

Maybe we can reduce the number of topics to see if RTD can handle that. Also, we can expand the documentation here and explain some of the parameters. For example, use frequency_threshold=0.001 to threshold the TFIDF for the term-based decoder and 0.05 for LDA-based decoders, and also how to use feature_group to select the labels from the annotation DataFrame.

@jdkent also suggested testing some of these non-running examples in the testing action (e.g., https://nimare.readthedocs.io/en/stable/auto_examples/01_datasets/02_download_neurosynth.html#sphx-glr-auto-examples-01-datasets-02-download-neurosynth-py)

[adelavega]

Ah, okay. That makes sense for testing, no need to have a large amount of topics. In a reaslitic user scenario more cores will be available.

Expanding the docs sounds good to me as well.