EulerMaruyama not implemented in V4?

Questions v5
1

Hi everyone,

it seems that EulerMaruyama raises “Not Implemented” in version 4. However, looking at the code, it seems that the code was ported from v3. Is this exception there because it does not work in v4 for some reason? If so, are there plans to fix it?

thanks!
tiago

2

Code was not ported. What you see is still the old V3 code.

Yes, we plan to fix it. I think there is a GSOC student working on it?

3

thanks! Although the code feels simple enough that it should run as is in pymc4, no? Is this something I can file a pull request on?

4

I think you can give it a try. There is a developer guide here: Implementing a Distribution — PyMC 4.1.3 documentation

It’s a bit more involved than it might seem from the outside, but certainly doable. Just keep that in mind.

5

Are there any news on this? I see the code remains unported and this is the one thing stopping me to switch v4…

6

I think @junpenglao was planning to finally refactor it, but he might not have the time :smiley:

7

Yes I am (slowly) getting to that.
@tpaixao do you have some example I can test out while porting the distribution?

8

What would you like to have? synthetic data generated from an SDE with known parameters to check if it recovers those?

9

Yep that would be useful!

10

Here are a couple of models, simulated and then estimated using pymc3.
Hope it helps. Let me know if you need anything else.

11

v4 port is working now, I am going to add a few more test but if you are inpatient: Refactor EulerMaruyama to work in v4 by junpenglao · Pull Request #6227 · pymc-devs/pymc · GitHub

1 Like
12

Outstanding! thanks a lot!