Dimensionality reduction in tica confusion

[eackad]

I've examined by tica results (out of 1992 dihedral features) and want to only keep the top 175. To do so, I'm trying to use the dim parameter:
nticas = 175 tica = pyemma.coordinates.tica(feats_orig, stride=1, lag=1,dim=nticas) tica_output = tica.get_output()
However, this does not seem to have an effect (or not the effect I expect). I would expect to now have 175 eigenvalues since I have 175 ticas I wanted to keep. However if I print(len(tica.eigenvalues)) I get 1992. But print(tica) gives
TICA(commute_map=False, dim=175, epsilon=1e-06, kinetic_map=True, lag=1,
ncov_max=inf, reversible=True, skip=0, stride=1, var_cutoff=0.95,
weights=None)
When I examine the tica_output object I find it only kept 175 of the features. I'd be great if someone could explain what the tica object is then? I can't find the docs on get_output() so I'm not clear on the difference.

[clonker]

As you already observed, the output is in the expected 175-dimensional space.

TICA is based on solving a generalized eigenvalue problem in the full space, meaning the decomposition contains all eigenvalues (up to a truncation threshold). The magnitude of the eigenvalues can be related to the timescales of processes, meaning that larger eigenvalues are 'slower' then smaller ones. The projection itself is then performed by using the top dim eigenvectors corresponding to the largest dim eigenvalues -- or rather, by evaluating the top dim eigenfunctions on the input data.

[clonker]

I am assuming this clarified it, closing the issue.