search

dasdibye / DL4RegAlloc

3 stars 1 forks source link

Implementing and understanding the csv files. #6

Open nahomtse opened 2 years ago

nahomtse commented 2 years ago

Hi,

I want to create a new dataset as input for the model instead of the training and test instances that were given. However to adjust my own data into type of yours I have to understand your data. The csv files column 01 until 201 consist of the edges. However could you explain the values listed in these columns since I can not understand what these values are (I also looked at the .c file but still don't understand).

Many thanks in advance!

dasdibye commented 2 years ago

Thanks for your interest Nahom. I am OOO till Sunday, 9th. I will get back to you after that.

From: Nahom Tsehaie @.> Sent: Tuesday, October 4, 2022 3:07 PM To: dasdibye/DL4RegAlloc @.> Cc: Subscribed @.***> Subject: [dasdibye/DL4RegAlloc] Implementing and understanding the csv files. (Issue #6)

Hi,

I want to create a new dataset as input for the model instead of the training and test instances. However to adjust my own data in to type of yours I have to understand your data. The csv files column 01 until 201 consists of the edges. However could you explain the numbers listed in these columns since I can not understand what these values are (I also looked at the .c file but still don't understand).

Many thanks in advance!

— Reply to this email directly, view it on GitHubhttps://github.com/dasdibye/DL4RegAlloc/issues/6, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AN6K56I7AZR2M7HC6ZY7ANTWBP3FJANCNFSM6AAAAAAQ4MLG4U. You are receiving this because you are subscribed to this thread.Message ID: @.***>

nahomtse commented 2 years ago

Hi, I solved this issue.

I have another question. Do you think there is a way to make this lstm network from supervised setting into semi-supervised setting by using graphs (in test and/or training data) which the chromatic number are unknown (which is actually the case in most of the time)? I was looking in a another loss function (vat loss), but could not implement in this system. Perhaps you know a certain (more easy) way.

Thanks in advance! Nahom

dasdibye commented 2 years ago

Hi Nahom-

Thanks for persisting with this and solving it. As I had taken leave and got caught up with stuff after returning I could not look into it. But anyway, in the csv file we create ONE bit set to TRUE/FALSE for each edge x->y that is present/not present. Since we support up to ~100 nodes, we allow up to 2 INT64 ( ie 128 bits ) that can map up to a max of 128x128 adjacency matrix. ( Yu would have probably figured this out by now since u solved the issue :-) )

Regarding your second question its a good one. We can probably use chromatic number and get some initial embedding for the nodes and then probably use something like a Transformer. But I havent tried it. What is VAT loss ?

-Thx Dibyendu