Questions about testing on TSPLIB
ucasyang opened this issue · comments
I have some questions about testing the performance on TSPLIB. When using Kool et.al's model (AM), I have trained the model on TSP100, and the testing performance on the original test dataset is perfect, but the testing performance (sample-1280) on berlin52 is so much worse (23441) than your listed result (8663) in the paper. Have you met this problem before? I have no idea what is the reason of this issue.
Hello,
Did you normalize coodinate of berlin 52 as [0,1] × [0,1] ?
Best,
Minsu Kim
-----Original Message-----From: ucasyang ***@***.***>To: alstn12088/LCP ***@***.***>Cc: Subscribed ***@***.***>Sent: 2022-09-08T15:04:16+09:00Subject: [alstn12088/LCP] Questions about testing on TSPLIB (Issue #1)
I have some questions about testing the performance on TSPLIB. When using Kool et.al's model (AM), I have trained the model on TSP100, and the testing performance on the original test dataset is perfect, but the testing performance (sample-1280) on berlin52 is so much worse (23441) than your listed result (8663) in the paper. Have you met this problem before? I have no idea what is the reason of this issue.
—Reply to this email directly, view it on GitHub #1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUTRVGUT2Z27ZVMT5RUAN4DV5F6WBANCNFSM6AAAAAAQHMEMQA.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
Thank you for replying! According to your suggestion, I have normalized the coordinate into [0,1] \times [0,1], but the distance is also decreased, namely 5.72...... Do you mean keep the maximum number of each dimension the same?
1. Normalize2. Solve the problem
3. Re-normalize tour length into original length.
…-----Original Message-----From: ucasyang ***@***.***>To: alstn12088/LCP ***@***.***>Cc: MinsuuuKim ***@***.***>, Comment ***@***.***>Sent: 2022-09-08T16:36:39+09:00Subject: Re: [alstn12088/LCP] Questions about testing on TSPLIB (Issue #1)
Thank you for replying! According to your suggestion, I have normalized the coordinate into [0,1] \times [0,1], but the distance is also decreased, namely 5.72...... Do you mean keep the maximum number of each dimension the same?
—Reply to this email directly, view it on GitHub #1 (comment), or unsubscribe https://github.com/notifications/unsubscribe-auth/AUTRVGVG2H35XKSXNPYKVR3V5GJQPANCNFSM6AAAAAAQHMEMQA.You are receiving this because you commented.Message ID: ***@***.***>