Octave returned: error: Two few unique points.

[alshedivat]

Octave returned: error: Two few unique points.

[hoangcuong2011]

Ah I forgot to mention that I got the error only when I initialized DKL's parameters using the network weights that I learned from a DNN:

model.load_weights('checkpoints/mlp_kin40k.h5', by_name=True)

If I don't do that, the training does not encounter the error at all. But of course training the whole model from scratch is not a good idea, as DKL usually provides a much worse performance without smart initialization.

Best,

[Yvonne-Ng]

Hi @hoangcuong2011
I have also ran into this problem and it seems to only appear whenever my layer have >=500 nodes I wonder if you ever figured out a work around for this? Thanks in advance.

[hoangcuong2011]

Hi,
I almost forget everything as it was already 1 year and a half since then.
But I recall that there was no workaround indeed (even with extensive googling). I had to give up with this as far as I remembered.

But If you happen to come up with any solution, please let me know as well.

Many thx!

[Yvonne-Ng]

Thanks! I will let you know if I learned anything