There were some problems with reproducing the results

[Qianbei-Liu]

I am interested in machine learning applied to logic synthesis, read your paper : SLAP: A Supervised Learning Approach for Priority CutsTechnology Mapping but I can't reproduce your results. I was intrigued by your ideas and appreciate your contribution.

there are some questions here :

1. There are three modified versions of abc(abc, abc-train, abc-train2) in the repository. Did you use abc to get the data comparison?

I can get the original data(ABC Original) listed in the table by using the normal version of abc form , but I can't get the same result by using the modified version of abc you provided.

The above is a normal adc, same as your experiment and your modified abc yields the following result

When I use the method in your paper to produce the adder_inf.txt of cut information, the results are also not the same as those in the paper as shown in the following figure

I think I've got some steps wrong or overlooked. Do you have any further information about SLAP?

2. Generate the data for training

The paper talks about ”use our modified version of ABC, which randomly shuffles the cut list on each node”. If it is random, the result will be different each time and the data(eg : adder_train_hashed.csv for 200 iterations)for training I got for the experiment was indeed different each time and did not seem to be pseudo-random. I think I got something wrong
Looking forward to hearing from you!

[ChulanZhang]

Dear Author @wlneto,
Appreciate your contribution to the community!

I got similar results. Could you help respond to this issue?

Hi @Qianbei-Liu ,
Have you figured it out?

Thanks!