Tensor Network models including MPS, 1d-SBS and CNN-PEPS for classification with PyTorch or TensorFlow 2 backend and optimizing MPO for Heisenberg XXZ model and 1d-Ising model with Density Matrix Renormalization Group (DMRG). We implement the tensor network algorithms with Google's tensornetwork.
(Green - Finished, Yellow - With Some Problems, Red - Not Finished)
Examples are in the root folder. They use PyTorch and TensorFlow as backends respectively.
For ground state solving with DMRG, you need to calculate the corresponding MPO (Matrix Product Operators) for target Hamiltonian. Here is an example:
In MPS for image classification, the pixels and labels are input to the MPS like this:
As for 1d-SBS, the structure is like this:
where the red nodes are "Copy Tensors". The dangling edges of 1d-SBS which denote the output are the inner product of the strings:
The structure of PEPS comes from Supervised Learning with Projected Entangled Pair States:
- You can choose to use
PyTorchorTensorFlow 2as backend. - Install
tensornetwork>=0.4.4,numpyandtqdm. - Clone this repo.
This is a very early version and it is unstable.