Setting random seeds for randomBN

[kenneth-lee-ch]

I tried using the following to fix the seed, but it doesn't seem to work for randomBN

import random
random.seed(20)

How can I fix a seed so that I can re-generate the same BN with the same CPT each time?

[phwuil]

Hi @kenneth-lee-ch , We have an internal random generator initialized with gum.initRandom() (https://pyagrum.readthedocs.io/en/1.6.1/miscellaneous.html#pyAgrum.initRandom) but it seems that gum.randomBN() does not use it ... :-(
We will have a look. Stay tuned! :-)

[phwuil]

Just to let you know the progress we made here :

• the aGrUm's generator was not completely used by some codes. Some times ago, we moved from rand() to std::mt19937 and this move was not complete.

• the BN generator has been more or less rewritten and is much faster now.

• But still we have to fix your main problem (setting random seed seems to work for basic random number generation but not for randomBN)

[kenneth-lee-ch]

@phwuil Thanks for the update!

[phwuil]

@kenneth-lee-ch : Fixed !

(tomorrow in pyAgrum-nightly , soon in 1.6.2)

[kenneth-lee-ch]

Also, thank you so much!

[kenneth-lee-ch]

@phwuil I just reinstall pyargrum-nightly and I ran into an error that didn't occur previously from the randBN() function

import pyAgrum as gum
graphnames=['X0', 'X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10', 'X11', 'X12', 'X13', 'X14', 'X15']
gum.randomBN(n=16, names=graphnames, ratio_arc=3)

[phwuil]

Hum, that's logical but I do not know how to prevent that : the new randomBN use the names X{i} in topological order ... So if you try to force the same names (using graphnames), you may try to change the name "X10" of a node whereas another has the same name.

Now you can just use : gum.randomBN(n=16,ratior_arc=3)

[phwuil]

PS- 16.2 will finally be 1.7.0 tomorrow

[phwuil]

PS2- Finally, I found a way to prevent these names collision. Will be in 1.7.0 tomorrow :-)

[kenneth-lee-ch]

Thanks! I look forward to the latest version!

…

On Thu, Mar 16, 2023, 6:59 PM Pierre-Henri Wuillemin < ***@***.***> wrote: PS2- Finally, I found a way to prevent these names collision. Will be in 1.7.0 tomorrow :-) — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AF3Z3CUGFGRWXDMAOVR3QJLW4OLLJANCNFSM6AAAAAAVNMKZ6Y> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[phwuil]

🤞 :-)

[phwuil]

1.7.0 is here !

[kenneth-lee-ch]

@phwuil It seems like I can no longer generate binary variables by having domain_size =2? I tried using domain_size =1, but it throws an error.

[phwuil]

:-) 1.7.1 is here !

[phwuil]

.. and by the way, I forgot in this CHANGELOG that I added a gum.lib.export.toFastBN :-)

[kenneth-lee-ch]

@phwuil When I use gum.initRandom(self.seed) along with gum.randomBN inside a function of a class. It seems like I cannot reproduce the graph for some reason.

[phwuil]

@kenneth-lee-ch , it seems to work here :

[kenneth-lee-ch]

@phwuil I didn't use a notebook but just compiled a .py file. Will that make any difference?

[phwuil]

It should not and it looks okt :

import pyAgrum as gum

class A:
  def f(self,seed):
    gum.initRandom(seed)
    return gum.randomBN(names="ABCDEFG")

a=A()
seeds=[5,4,0,5,4]
for seed in seeds:
  print("************")
  print(f"** {seed=}")
  print("************")
  bn=a.f(seed)
  print(f'  => {bn.parents("G")=}')
  print(bn.cpt("A"))

run :

[phwuil]

can you share a minimal example of your problem ?

[kenneth-lee-ch]

I think it's a problem on my end. Thank you so much!