How to use mps_decomposition? about tensorly HOT 4 CLOSED

Hi, @Yolanpera. It seems to me that the last update of tensorly on pypi.org was on the 8th of May 2018. Whereas module mps_decomposition was added sometime in July. So your best option would be to install from the source code available in this repository:

# clone the repository and cd into it
git clone https://github.com/tensorly/tensorly
cd tensorly

# Install it in your environment
pip install .

Then you could do something like

import numpy as np
import tensorly as tl
from tensorly.decomposition import matrix_product_state

tensor = tl.tensor(np.random.random((10, 10, 10)))

factors = matrix_product_state(tensor, rank=(1, 10, 10, 1))

tensor_rec = tl.mps_to_tensor(factors)
rec_error = tl.norm(tensor - tensor_rec, 2)/tl.norm(tensor)

print("="*20)
[print(f.shape) for f in factors]

print("="*20)
print(tensor_rec.shape)

print("="*20)
print(rec_error)

As far as documentation is concerned, you have been looking at the pages related to the stable version of tensorly, which, I assume, corresponds to the latest version of tensorly on pypi.org. In order to see the most recent API reference, you need to open http://tensorly.org/dev instead. By looking at the continuous integration scripts, each time the master branch of tensorly/tensorly successfully passes unit tests, the latest version of the website is deployed to tensorly/tensorly.github.io as well.

from tensorly.

Hi,@IlyaKisil Thank you very much!
It works for me!

from tensorly.

I'll update the pypi version.

from tensorly.

Done - you can upgrade to 0.4.3 with pypi or conda.

from tensorly.