Add get_tensor_network_state method #5
Conversation
At the moment, this changes nothing. It will, in the future, to remember that those attributes actually belong to TensorNetwork and not to SimpleUpdate. If someone changes them from the TensorNetwork attribute, these will also change. The current structure reflects this fact.
Moved methods that absorb weights into TensorNetwork class. Created `TensorNetwork .get_tensor_network_state()` method
|
Thank you very much for putting this together, it looks very good. |
|
I've responded to the comments above. Tell me if you wish to change something. |
|
It all looks very good. Just a final question, did you had the chance executing the test module, just to check that everything still runs properly? If not, could you do that? just run |
+ run all tests in `test_main`
I pass all tests besides |
Good enough! |
In response to issue #4:
Main changes:
SimpleUpdateclass are now properties, using the@propertydecorator.TensorNetworkclass. This supports the following point.get_tensor_network_state()which returns a copy of theTensorNetworkobject, with weights absorbed into neighboring tensors, to reflect the actual state of the systemafh_peps_ground_state_experiment()instead of saving the original TensorNetwork, now the return value ofget_tensor_network_state()is usedTensorNetworkandSimpleUpdateclasses. Mostly to annotate return values to allow linters to provide coding-hints for users, according to PEP 484.Possible disagreements or issues:
The return type of
get_tensor_network_state()is currently aTensorNetworkobject.It might make more sense to return a simpler list of tensors (i.e., type
list[np.ndarray]), where the list is simplyself.tensors.