HTBB

Hierarchical Tucker for Black Box gradient-free discrete approximation and optimization HTBB

Installation

1. Install anaconda package manager with python (version 3.8);

2. Create a virtual environment:

   conda create --name htbb python=3.8 -y

3. Activate the environment:

   conda activate htbb

4. Install dependencies:

   pip install teneva_opti==0.5.3

   When update teneva_opti version, please, do before: pip uninstall teneva_opti -y.

5. Install dependencies for benchmarks (it is optional now, since we run only analytic functions!):

   wget https://raw.githubusercontent.com/AndreiChertkov/teneva_bm/main/install_all.py && python install_all.py --env htde

   You can then remove the downloaded file as rm install_all.py. In the case of problems with scikit-learn, uninstall it as pip uninstall scikit-learn and then install it from the anaconda: conda install -c anaconda scikit-learn.

6. Optionally delete virtual environment at the end of the work:

   conda activate && conda remove --name htbb --all -y

Computations

1. Run the approximation problems as:

   python run_func_appr.py

   The results (for $d = 256$) will be in the result_func_appr folder. You can use the flag --show to only present the saved computation results. For the case of higher dimensions (d = 512 and d = 1024) we saved the results in the result_func_appr_d[d] folder. To show the results, please, run the script like python run_func_appr.py --show --fold result_func_appr_d512 --without_bs.

2. Run the optimization problems as:

   python run_func_opti.py

   The results will be in the result_func_opti folder. You can use the flags --with_no_calc to only present the saved computation results.

Authors

• Gleb Ryzhakov (Basic ideas; raw code for proof of concept)
• Andrei Chertkov (Code speed rewriting & speed up; most of experiments; checking)
• Artem Basharin (PEPS part; testing)
• Ivan Oseledets (Supervision)

---

✭__🚂 The stars that you give to HTBB, motivate us to develop faster and add new interesting features to the code 😃