tn-tools

tn-tools is an assortment of miscellaneous tools and utilities that I use when developing tensor network algorithms in Python3. Some of these are pretty specific to my own uses, such as my favourite way of formatting log messages. The ones most useful to others would be

• datadispenser.py (which uses pact.py), an automated disk-storage and retrieval system for data generated by algorithms. Meant to automate the process of "did I already run this algorithm with these parameters, and if I did, where did I store the output?"
• ncon_sparseeig.py, that implements sparse decomposition routines for tensor networks with an NCon-like syntax.

The level of documentation and code beauty varies. datadispenser is pretty decent, ncon_sparseeig intimidates even me. The tools here rely on the abeliantensors and ncon packages.

Installation

pip install --user git+https://github.com/mhauru/tntools

The files

pact.py A module for storing data on the disk, such that each piece of data is uniquely identified by a name for the type of data, and a dictionary of parameters. For instance, the name could be "MERA_disentangler" and the parameters would be all the parameters that went into producing this disentangler, such as the bond dimension, the physical model it's used for and the optimization method. pact can then store the data with the identifying information, and fetch the data given the identifying information.

datadispenser.py A module that generates data using various algorithms, and stores the data on the disk (using pact). The idea is that a user just tells datadispenser what kind of data she wants ("I want a MERA disentangler produced using these parameters"), and datadispenser either finds the data already on disk, or generates it for the user, storing it as well, in case it is requested later. It also generates any prerequisite data necessary (such as the other tensors in the same MERA).

ncon_sparseeig.py A module that implements two user-facing functions: ncon_sparseeig and ncon_sparsesvd. They provide a convenient interface, similar to that of the ncon package, for doing eigenvalue and singular value decompositions of tensor networks using power methods (from scipy.sparse.linalg), without ever contracting the full network.

multilineformatter.py A formatter class for the Python logging module that formats multiline messages nicely.

logging_default.conf A default configuration for the Python logging module that for instance uses the above multilineformatter.

initialtensors.py A module that produces initial tensors for various lattice models in 2D and 3D, that can be used to write down the partition function of the model. Used as starting points for many tensor network algorithms.

initialtensors_setup.py A module for interfacing initialtensors.py with datadispenser.

modeldata.py A module that provides exact data for some solvable lattice models, such as exact free energies and CFT data. Handy for benchmarking.

yaml_config_parser.py A module for reading in parameters for various programs in the YAML format. Supports both .yaml files as configuration files, and appending/overriding parameters using command line arguments.