LowPopArt

This is the code for the paper 'Efficient Low-Rank Matrix Estimation, Experimental Design, and Arm-Set-Dependent Low-Rank Bandits'

Required software

• python 3 with numpy, scipy, sklearn, cython, ipdb, cvxpy, tqdm, mosek
• Especially about mosek, you need an additional license file to actually use the package.

Compile needed

• in matrixrecovery, run cython myutils_cython.pyx and then python3 setup.py install
• for mac
  • in pyOptSpace_py3_custom, run cython optspace.pyx then python3 setup.py install
  • copy the created *.so file to the upper directory
• for linux
  • do the same as mac above, but in the directory pyOptSpace_py3_linux_custom

To replicate the Figure 1 (left) in the paper

• Run 'Figure1_left_estimator_experi_easy.ipynb' in 'Figure 123'

To replicate the Figure 1 (right) in the paper

• Run 'Figure1_right_estimator_experi_hard.ipynb' in 'Figure 123'

To replicate the Figure 2 (left) in the paper

• Run 'Figure2_left_ETC_regret_analysis.ipynb' in 'Figure 123'

To replicate the result in Figure 2 (right) in the paper

1. Run the script ./running_script.sh
2. Move all .pkl files to a subfolder in /res-20230928/R1_lambda_1/T100000 folder
3. remove all the date prefix on .pkl file name. For example, change the name '20230928Sun-123456-bmoracle.pkl' to 'bmoracle.pkl'
4. Run python3 analyze_expr01_230928_paper.py