Meirinhos

> > a downside is that uses the FastTransforms > > I'm not sure this is desirable. I have now removed the need for FastTransforms of the Cauchy integrator. The...

That is an excellent transformation which I unfortunately did not know about. I still had the hope that the modified Clenshaw-Curtis quadrature method could outperform quadgk for nastier single-poled integrals...

I was under the impression that inplace plans would compute the transforms in a pre-determined location (since I thought FFT operations could not actually be done in place!). For repeated...

I only glanced at the paper and the formula for the angle should actually be `arccos((u . v)/(|u| |v|))`. Do you have any minimal example you can share with us?...

Building on @LRydin's answer, note that a power `[m, n]` enters the formula, as in the paper's Data Analysis, as ``` D_{x^m y^n} = \int (x - x')^m (y -...

It just occurred to me that `angle_between` may not be 100% suitable since it's just the angle magnitude, defined in `[0, pi]`, and since we are adding angles, it would...

Cheers, @sshkhr I have tested the optimiser for PyTorch 1.3 and it seemed to work fine. It just doesn't support parameter groups (so optimising for different parameters with different optimiser-hyper-parameters...

The CNN architecture you posted should work just fine, @opooladz. See the `hf_test.py` file for how to apply the optimiser to a `torch.nn.Module`