fabio-amadio/cgpdm_lib

Controlled Gaussian Process Dynamical Models (CGPDMs) Python library.

cgpdm_lib's Issues

Loss calculate issues

Hi & thanks for your great work.

I would like to reference the loss calculated here:

cgpdm_lib/cgpdm_lib/models.py

Lines 465 to 467 in 712600d

	return self.D / 2 * log_det_K_y + 1 / 2 * \
	torch.trace(torch.linalg.multi_dot([Ky_inv, Y, W2, Y.transpose(0,1)])) \
	- N * log_det_W

It seems the loss_y calculation omits parts of the terms in eqn.16 in Wang et al. (2005), that is $\sum_{j} \ln \beta_{j}$ . Same as the loss_x.

Or is this part implicitly included in the previous calculation?

Add 120 deg DATA

Add 120 deg trajectories to DATA folder and update test scripts.