Strange behaviour when testing with constant values about matrixprofile-ts HOT 8 CLOSED

@dvdcoliveira what version of the library are you using? Is it the latest (0.0.8)?

from matrixprofile-ts.

Hi! @vanbenschoten,

I was using python version 3.7.1 and matrixprofile-ts version 0.0.6. I updated to 0.0.8 but the problem still persists.

If I use scrimp++ this does not happen.

Additional information:

• numpy version 1.15.4
• Windows version 10 Enterprise.

Many thanks for the support,
David

from matrixprofile-ts.

Hi,
Found the same issue when appending a constant value zero to the rawdata example.
Solve it by adding a very small noise to the example pattern:
N = 1046
ix = np.arange(N)
signal = np.sin(2*np.pi*ix/float(32))*0.01
m = 32
mp = matrixProfile.stomp(pattern+signal,m)

Probably it's a problem with the time series z normalization. When I run the matrixprofile with the constant values I get the following errors:

• matrixprofile\utils.py:177: RuntimeWarning: divide by zero encountered in true_divide
• utils.py:200: RuntimeWarning: divide by zero encountered in true_divide
• distanceProfile.py:126: RuntimeWarning: invalid value encountered in sqrt
• matrixProfile.py:249: RuntimeWarning: invalid value encountered in less idsToUpdate = distanceProfile < mp

Hope it helps,
Alex

from matrixprofile-ts.

@AlexandreRodrigues1 thanks much for the contribution! @aouyang1 did you ever figure out a solution to this challenge for go-matrixprofile?

from matrixprofile-ts.

Thank you to @vanbenschoten and the other contributors too, for the excellent job done in implementing the Matrix Profile algorithms in python :)

from matrixprofile-ts.

I'll need to test this case as well. I don't believe the golang version handles this nicely either. Hopefully can get back to you all on this in the next few days

from matrixprofile-ts.

This is the same as #11

from matrixprofile-ts.

Closing as issues is same as #11

from matrixprofile-ts.