genetproject/genet

The repository of Genet project.

genet's Issues

reward calculation bug in fig_reproduce.sim.rl_test_clean*.py

There are some bugs in fig_reproduce.sim.{rl_test_clean_udr_1.py,rl_test_clean_udr_2.py,rl_test_clean_udr_3.py,rl_test_clean_udr_real.py}. The original function of given_string_mean_reward is:

def given_string_mean_reward(plot_files ,test_dir ,str):
    matching = [s for s in plot_files if str in s]
    reward = []
    count = 0
    for log_file in matching:
        count += 1
        # print(log_file)
        with open( test_dir + '/' + log_file ,'r' ) as f:
            for line in f:
                parse = line.split()
                if len( parse ) <= 1:
                    break
                reward.append( float( parse[6] ) )  # bug here, index of reward column is 7
    print( count )
    return np.mean( reward[1:] )  # bug here, if we want to skip the first reward, we should skip for every log file.

After fixing the bugs, this function should be:

def given_string_mean_reward(plot_files ,test_dir ,str, skip_first_reward=True):
    matching = [s for s in plot_files if str in s]
    reward = []
    count = 0
    for log_file in matching:
        count += 1
        # print(log_file)
        first_line = True
        with open( test_dir + '/' + log_file ,'r' ) as f:
            for line in f:
                parse = line.split()
                if len( parse ) <= 1:
                    break
                if first_line:
                    first_line = False
                    if skip_first_reward:
                        continue
                reward.append( float( parse[7] ) ) 
    print( count )
    return np.mean( reward )

Vivre sans temps mort. (Live without wasted time.)

Friend,

There is never enough time, thank you for yours. Thank you for your integrity. Thank you for your humility. Thank you for your presence. I would appreciate it if you could help me get past this issue: Mac OSX, not supported!. Do you have a work around?

V.R.
E. M. Gertis
www.evan-gertis.com

question

I am sorry to interrupt you,i am very interesed in your paper,but i meet a problem,I can not open this link---https://genet.blob.core.windows.net/sigcomm22/models.tar.gz.

Train Pensieve with Genet on FCC/Norway traces

Hi!

First of all, great work!

I'd like to experiment with Genet and train a few other variants of Pensieve.
Concretely, I'd like to train Pensieve with some of the FCC and Norway traces.

As far as I can see, this is not supported by the current code, correct?
I found this file, which I believe does the Genet training for Pensieve. However, currently, it only works with synthetic traces.

First of all: is my understanding correct? Or is there some other code that can train Pensieve with other traces and I just haven't found it yet?

If there isn't, I considered modifying this part of the code, i.e. the black_box_function, to load existing traces instead of generating them. Would that make sense?

Is there any data description?

Hi,

I'm trying to understand the real-life data used by Genet and any help would be appreciated.

Where can I find a technical description of the real-life traces, i.e. the FCC and Norway traces?

Some of these files only have one value per line, others two. Is this explained somewhere?

Furthermore, how are these traves generated from the raw FCC/Norway data (which looks quite different than the traces).

Thank you!

Problems trying to run the evaluation with Mahimahi.

Happy new year, everyone!

I'm trying to run some parts of Genets evaluation, specifically the emulated part with Mahimahi.

I followed the installation instructions, but cannot get Mahimahi to run, it always crashes with some permission errors.
I've created an issue in the Mahimahi repo with more details, but I was wondering if you had similar issues with Mahimahi?

Is there any specific configuration/setup I need to to to make it work?