Use MixMatch on tabular data. about mixmatch

[david-berthelot]

You can use it without augmentation. MixUp, which is baked in the method, might be sufficient for your data.

[Julianbaozi]

Thank you for your response. That is very helpful.

…

On Sat, Jun 20, 2020 at 2:43 PM David Berthelot ***@***.***> wrote: You can use it without augmentation. MixUp, which is baked in the method, might be sufficient for your data. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/google-research/mixmatch/issues/28*issuecomment-647048606__;Iw!!Mih3wA!TG6N_UE0FvYQ36Wv2Q6F0e0qK32wmzHmFoOAbgYemNlw7aoBZban4naHl6tE_g$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AKUX66652EGNDCWAVYHVISLRXUUONANCNFSM4ODQHNVQ__;!!Mih3wA!TG6N_UE0FvYQ36Wv2Q6F0e0qK32wmzHmFoOAbgYemNlw7aoBZban4nZULzvxNg$> .

[Julianbaozi]

Should I do cross validation? I have only 360 labeled points. The validation set should only take labeled data and would have around 90 points (4 folds). I'm I correct?

[david-berthelot]

You'll need a test set (or take from your labeled examples to make on if you don't have one) to assess the performance of the model. That's always a requirement. Now as to the proportion to keep for training and for testing, I don't know what it should be. I would experiment with various values (say 50/50, 75/25, and so on) and pick the smallest test set that works.

Hello, it's not that semi supervised learning like miamache can't be used on unbalanced data sets. I use it on my own data set. The test set is accurate, and the test set rises by 2 epochs and then decreases continuously. However, the loss of training set is declining. What's the matter? Is my dataset too unbalanced

[Julianbaozi]

I think your test set might be biased. Or you can try to fine tune the parameters

…

On Wed, Dec 9, 2020 at 5:08 PM Wan-Dou ***@***.***> wrote: Hello, it's not that semi supervised learning like miamache can't be used on unbalanced data sets. I use it on my own data set. The test set is accurate, and the test set rises by 2 epochs and then decreases continuously. However, the loss of training set is declining. What's the matter? Is my dataset too unbalanced — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/google-research/mixmatch/issues/28*issuecomment-742165770__;Iw!!Mih3wA!TtD_Q0Ir1c5jUXosG0Mekad8q1Fv-b2eXfATSmQDaokaiTZ2PFIyuFqaqwfyoA$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AKUX664EVY4OO7CUEWFOB63SUANP5ANCNFSM4ODQHNVQ__;!!Mih3wA!TtD_Q0Ir1c5jUXosG0Mekad8q1Fv-b2eXfATSmQDaokaiTZ2PFIyuFrUcle4Tw$> .