search

benchopt / benchmark_tv_1d

TV Denoising in 1D
2 stars 7 forks source link

add blocked dataset #22

Closed EnLAI111 closed 2 years ago

EnLAI111 commented 2 years ago

@EnLAI111 can you share a figure when you have one? 🙏

Several algo don't converge to the same point. I have to check out what's wrong. image

agramfort commented 2 years ago

looks great!

Message ID: @.***>

EnLAI111 commented 2 years ago

A few more comments. but overall, the results look pretty nice.

I agree that this is surprising that the values do not converge to the exact same spot.. any idea why? is this only for splitting algorithm? if yes, this could be due to the need for a scaling of the multiplier?

The problem of convergence appears for all the algorithms of duality when A is random, including CondatVu, ADMM, Chambolle-Pock and Dual prox gradient descent. When the signal is block, the problem is more evident then that when signal is sinusoidal.

tomMoral commented 2 years ago

Them, maybe we have an issue on how we compute the primal from the dual variable?

tomMoral commented 2 years ago

The CI s finally green :green_circle: :grin: