The results of different runs differ a lot.

[klovbe]

I am implementing this code on my biological datasets. When I run the code multiple times, it produces quite different results. The ARIs vary from 0.4 to 0.8. I find out that the pretrained model matters a lot to the results. How can I achieve robust results by getting a good pretrained model? Is the pretrained denoising autoencoder important to stacked autoencoder when the structure of the dae is not exactly the same as that of sdae?

[eelxpeng]

What code are you using? stacked denoising autoencoder?

[klovbe]

Sorry about the ambiguity! When I run the code VaDe multiple times, it produces quite different results. The ARIs vary from 0.4 to 0.8. I find out that the pretrained sdae model results matters a lot to the VaDe clustering results. How can I achieve robust results by getting a good pretrained sdae model? Is the pretrained denoising autoencoder important to sdae when the structure of the dae is not exactly the same as that of sdae?

[eelxpeng]

This is actually the issue of the original paper and their released code. I also find this issue annoying. But I find that using xavier initialization for the weights could alleviate the problem a lot. You could try.

[chulaihunde]

How to get the pretrained_vade-3layer.pt ?

[eelxpeng]

@chulaihunde I converted from original VaDE page. But I really don't know how they get the pretrained weights. Let me know if you figured that out.

[chulaihunde]

May be able to refer to https://github.com/piiswrong/dec/blob/master/dec/pretrain.py