wangsix / vmo Goto Github PK

I find the name data to be confusing in the oracle's class attributes.
I'd rather go for symbols instead, to clearly distinguish it from the concrete data, the features. With an -s, because there are several possible symbols accessible for a given state!

And rename f_array to feature at the same time, because, who knows, a feature doesn't necessarily have to be an array! And anyway, the interface should be kept separate from the actual implementation, so mentioning arrays is bad practice.

@wangsix Both demo notebooks use direct accesses to VMO.latent. Furthermore, this method is not documented in the class' definition.

Consider creating a method to access this attribute if it is really needed, in order to respect the encapsulation principle.

Given

SYMBOLIC_SEQ = [0, 0, 1, 0, 2, 0, 0, 1, 0, 1, 0, 2, 0, 0, 1]
p = build_oracle(SYMBOLIC_SEQ, 'f')
print(vmo.generate.improvise(p, 5))

I get

Traceback (most recent call last):
  File "test.py", line 9, in <module>
    print(vmo.generate.improvise(p, 15))
  File "/home/jcelerier/score/base/addons/score-addon-vmo/3rdparty/vmo/vmo/generate.py", line 87, in improvise
    s.append(improvise_step(oracle, k, LRS, weight))
  File "/home/jcelerier/score/base/addons/score-addon-vmo/3rdparty/vmo/vmo/generate.py", line 47, in improvise_step
    trn_link = [s + 1 for s in oracle.latent[oracle.data[i]] if
TypeError: only integer scalar arrays can be converted to a scalar index

I changed the function to the following to make it work (just a cast to int for the oracle data sequence):

    latent = oracle.latent[int(oracle.data[i])]
    if prune:
        prune_list = range(i % prune, oracle.n_states - 1, prune)
        trn_link = [s + 1 for s in latent if
                    (oracle.lrs[s] >= lrs and
                     (s + 1) < oracle.n_states) and
                    s in prune_list]
    else:
        trn_link = [s + 1 for s in latent if
                    (oracle.lrs[s] >= lrs and (s + 1) < oracle.n_states)]
    if not trn_link:
        if i == oracle.n_states - 1:
            n = 1
        else:
            n = i + 1
    else:
        if weight == 'lrs':
            lrs_link = [oracle.lrs[s] for s in latent if
                        (oracle.lrs[s] >= lrs and (s + 1) < oracle.n_states)]
            lrs_pop = list(itertools.chain.from_iterable(itertools.chain.from_iterable(
                [[[i] * _x for (i, _x) in zip(trn_link, lrs_link)]])))
            n = np.random.choice(lrs_pop)
        else:
            n = trn_link[int(np.floor(random.random() * len(trn_link)))]
    return n

@wangsix : the MO class offers a parameter suffix_method (currently with two recognized values,
'inc' and 'complete'), VMO and FO don't support it.
Where is this parameter introduced? Is it specific to the repository or does it appear in an article?

Is there any reason why MO should benefit from it and not FO/ VMO?
I am rewriting all three classes to use a unified add_class method, and shall integrate
this parameter regarding the suffix links within this generic function in case it's relevant.

Hello,
when going to the http://chengiwang.com/ website, it's only a "default" page.

Where could we get more informations on VMO ?

Best,

Hi,

I saw that vmo implements the algorithm found at

Lefebvre, A., Lecroq, T., & Alexandre, J. (2003). An Improved Algorithm for Finding Longest Repeats with a Modified Factor Oracle. Journal of Automata, Languages and Combinatorics, 8(4), 647–657.

However, the factor oracle for the string abbcabcdabc built by vmo does not correspond to the shown in the paper (figure below). VMO returns lrs[11] = 2 and sfx[11] = 4

Cheers,
Jaime.

I have a request which would greatly help me for project on which I am curently working.
I would be interested to have a code exemple which uses Markov Oracle in VMO, similar to the one you use in your publication (A Novel Sequential Data Points Clustering algorithm with aplication to 3D Gesture) into the experiments sections.

Thanks.

The codes for classes FactorOracle, MO and VMO are all very close and share many identical lines, making it hard to uniformly update it.
This could probably be fixed by refactoring the code.

I think having (for all newly created oracle) oracle.symbol[0] == None rather than oracle.symbol[0] == 0 would make more sense.
What do You think about it?

Given this code :

from vmo.VMO.oracle import build_oracle
import vmo.generate

SYMBOLIC_SEQ = [0, 0, 1, 0, 2, 0, 0, 1, 0, 1, 0, 2, 0, 0, 1]
p = build_oracle(SYMBOLIC_SEQ, 'a')
print(vmo.generate.improvise(p, 5))

I get (all the time)

vmo/VMO/oracle.py:567: RuntimeWarning: invalid value encountered in less
  I = np.where(dvec < self.params['threshold'])[0]
[1, 2, 3, 4, 5]

what should I do to get correct improvisation ? thanks!

Hello,

It would be interesting to have the python notebooks (VMO tutorials) in binder