1- Check the value of self.M_ULA in environment.py. Then create a subfolder with that value and run the main.py file. For example figures_M=4 if self.M_ULA = 4. Move the measurement_x_seedy.txt files to that folder.
2- Change self.M_ULA to the next value (I did 4, 8, 16, 32, 64). Then create another subfolder with that name: figures_M=8, ... Move the measurements to this new folder.
3- Repeat step 2 until you are done with the number of M's.
4- Copy parse_new.py to every one of these figures_M=k folders. Then run it from the shell in from these folders: ./parse_new.py. This will generate 4 files in each of these folders.
5- In main.py uncomment lines 426 and 435. Change the seed to reflect the seed of the desired convergence reward. Comment 437 and 440. This now allows the optimal algorithm to run. Change self.M_ULA as in step 2 above. Store results to the folder 'figures M=k optimal' where k = 4, 8, ..., 64. When these are created, you still need to run parse_new.py in each one of these folders similar to step 4 above.
6- Make sure a folder 'figures' is created. Change line 20 in plotting_v6.py to reflect the path where your 'figures' folder is. Now run plotting_v6.py and you should be good.
Due to the various library version dependencies and the reproducibility issues with TensorFlow 1.8 on GPU, I cannot guarantee that the output you would get running this code would match my outputs.