Could you tell me how you use T-SNE? Thank you!

[sunhappy6900]

Could you tell me how you use T-SNE? Thank you!

[bbuing9]

Hi, T-SNE plots (in Figure 4) are obtained as follow:

1. Training the network with the desired method under the target imbalanced ratio, e.g., M2m with \rho=100

2. Then, randomly select N training samples per class which will be plotted using T-SNE. (E.g., N=50 in the paper)

3. Obtaining the "penultimate features", i.e., before the linear classifier, of the selected samples using the trained network

4. The, plot the obtained features using any T-SNE code, e.g., Multicore-TSNE (https://github.com/DmitryUlyanov/Multicore-TSNE)

Best,
Jaehyung

[sunhappy6900]

Thank you very much!

…

------------------ 原始邮件 ------------------ 发件人: "alinlab/M2m" ***@***.***>; 发送时间: 2021年7月18日(星期天) 中午1:18 ***@***.***>; ***@***.******@***.***>; 主题: Re: [alinlab/M2m] Could you tell me how you use T-SNE? Thank you! (#7) Hi, T-SNE plots (in Figure 4) are obtained as follow: Training the network with the desired method under the target imbalanced ratio, e.g., M2m with \rho=100 Then, randomly select N training samples per class which will be plotted using T-SNE. (E.g., N=50 in the paper) Obtaining the "penultimate features", i.e., before the linear classifier, of the selected samples using the trained network The, plot the obtained features using any T-SNE code, e.g., Multicore-TSNE (https://github.com/DmitryUlyanov/Multicore-TSNE) Best, Jaehyung — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.