Error while calculating VAMP2 score at different lag times

[Bazzinga18]

Hii, While VAMP2 score analysis for a particular feature by varying the dimension parameter for several lag times i got this error: RuntimeError: requested more output dimensions (2) than dimension of input data (1)

although total TICA dimension is 220

i used this particular code for the same:

def score_cv(data, dim, lag, number_of_splits=5, validation_fraction=0.5):
    with pyemma.util.contexts.settings(show_progress_bars=False):
        nval = int(len(data) * validation_fraction)
        scores = np.zeros(number_of_splits)
        for n in range(number_of_splits):
            ival = np.random.choice(len(data), size=nval, replace=False)
            vamp = pyemma.coordinates.vamp(
                [d for i, d in enumerate(data) if i not in ival], lag=lag, dim=dim)
            scores[n] = vamp.score([d for i, d in enumerate(data) if i in ival])
    return scores
lags = [1, 2, 5, 10, 20]
dims = [i + 1 for i in range(10)]

fig, ax = plt.subplots()
for i, lag in enumerate(lags):
    scores_ = np.array([score_cv(data, dim, lag)for dim in dims])
    scores = np.mean(scores_, axis=1)
    errors = np.std(scores_, axis=1, ddof=1)
    color = 'C{}'.format(i)
    ax.fill_between(dim, scores - errors, scores + errors, alpha=0.3, facecolor=color)
    ax.plot(dim, scores, '--o', color=color, label='lag={:.4f}ns'.format(lag * 0.4))
ax.legend()
ax.set_xlabel('number of dimensions')
ax.set_ylabel('VAMP2 score')
fig.tight_layout()

Please let me know as soon as possible;) thanks

[clonker]

Hi Bazzinga! To see what is going on we are going to need a bit more context. It seems to me though that you are giving it not the TICA-transformed data with 220 dimensions but something else. You can embed code fragments by embracing it with triple backticks like ``` by the way :slightly_smiling_face:

[Bazzinga18]

hey thanks for your response. Let me show you the script from starting please tell me what's wrong:

feat.add_distances(indices=(1969,1979,1989,2004,2028,2043,2050,2066,2083,2102,2121,2143,2165,2179,2189,2204,2218,2237,2249), indices2=(2005,2029,2044,2051,2067,2084,2103,2122,2144,2166,2180,2190,2205,2219,2238,2250,2265,2275,2285), periodic=False)
data = pyemma.coordinates.load(files, features=feat)
labels = ['atom\ndistances']
lags = [1, 2, 5, 10, 20]
dims =  [i + 1 for i in range(10)]

fig, ax = plt.subplots()
for i, lag in enumerate(lags):
    scores_ = np.array([score_cv(data, dim, lag)for dim in dims])
    scores = np.mean(scores_, axis=1)
    errors = np.std(scores_, axis=1, ddof=1)
    color = 'C{}'.format(i)
    ax.fill_between(dim, scores - errors, scores + errors, alpha=0.3, facecolor=color)
    ax.plot(dim, scores, '--o', color=color, label='lag={:.4f}ns'.format(lag * 0.4))
ax.legend()
ax.set_xlabel('number of dimensions')
ax.set_ylabel('VAMP2 score')
fig.tight_layout()

anything else you need to know?

[clonker]

Thanks! Can you please post the output of feat.describe()?

[clonker]

For scoring you can also try the following and see if it works, but I suspect the problem is elsewhere:

from deeptime.decomposition import VAMP, vamp_score_cv

estimator = VAMP(lagtime=lag, dim=dim)
scores = vamp_score_cv(estimator, data, 
    lagtime=1000  # note that this is a different lagtime: it is used to split the trajectory into blocks of length "lagtime" or longer
)

[Bazzinga18]

Thanks! Can you please post the output of feat.describe()?

feat.describe()[:10] ['DIST: ASP 120 O 1969 0 - ARG 124 N 2005 0', 'DIST: ASP 120 O 1969 0 - GLU 125 N 2029 0', 'DIST: ASP 120 O 1969 0 - GLY 126 N 2044 0', 'DIST: ASP 120 O 1969 0 - VAL 127 N 2051 0', 'DIST: ASP 120 O 1969 0 - MET 128 N 2067 0', 'DIST: ASP 120 O 1969 0 - LEU 129 N 2084 0', 'DIST: ASP 120 O 1969 0 - ILE 130 N 2103 0', 'DIST: ASP 120 O 1969 0 - LYS 131 N 2122 0', 'DIST: ASP 120 O 1969 0 - LYS 132 N 2144 0', 'DIST: ASP 120 O 1969 0 - THR 133 N 2166 0']

feat.dimension() 361

[clonker]

Alright that is good :slightly_smiling_face: can you please check the contents of data? It should be a list of length "number of files" with each list element a numpy array of shape (n_frames, 361).

[Bazzinga18]

type of data: <class 'numpy.ndarray'> lengths: 25420 shape of elements: (361,) n_atoms: 2399

[clonker]

alright so this means that you don't have several trajectory flies but just one, is that right? in that case the function you are using is not appropriate to compute a cross-validated vamp score. please use the one I suggested above:

from deeptime.decomposition import VAMP, vamp_score_cv

estimator = VAMP(lagtime=lag, dim=dim)
scores = vamp_score_cv(estimator, data, 
    lagtime=1000  # note that this is a different lagtime: it is used to split the trajectory into blocks of length "lagtime" or longer
)

[Bazzinga18]

yes there is only one trajectory file. okay thanks i tried to do that. :)

[clonker]

so... did it work? :wink:

[clonker]

Please feel free to reopen if this is still an issue