[theano-users] ERROR (theano.gof.opt): EquilibriumOptimizer max'ed out by 'cut_gpua_constant_transfers'

[Ioannis Marras]

Hi all,

After the compilation of the training and validation function of my CNN I
am getting this error:

ERROR (theano.gof.opt): EquilibriumOptimizer max'ed out by
'cut_gpua_constant_transfers'. You can safely raise the current threshold
of 5.000000 with the theano flag 'optdb.max_use_ratio'.

Although my CNN was running after this error I am wondering if I can ignore
this error or not....


What does this error means? (I found on the internet that this is related
to T.set_subtensor(). In my code there are repeated applications of
T.set_subtensor())

Many Thanks,

Ioannis

[Frédéric Bastien]

The function will run correctly after this. So you use it. But it could run
less efficiently. Can you try this Theano flag:

optdb.max_use_ratio=50

And tell us if this fix your problem?

Fred

Post by Ioannis Marras
Hi all,
After the compilation of the training and validation function of my CNN I
ERROR (theano.gof.opt): EquilibriumOptimizer max'ed out by
'cut_gpua_constant_transfers'. You can safely raise the current threshold
of 5.000000 with the theano flag 'optdb.max_use_ratio'.
Although my CNN was running after this error I am wondering if I can
ignore this error or not....
What does this error means? (I found on the internet that this is related
to T.set_subtensor(). In my code there are repeated applications of
T.set_subtensor())
Many Thanks,
Ioannis
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an
For more options, visit https://groups.google.com/d/optout.

[Ioannis Marras]

Hi Fred,

Many thanks for your help.

Using this Flag, this error disappeared.

Best,

Ioannis

Post by Frédéric Bastien
The function will run correctly after this. So you use it. But it could
optdb.max_use_ratio=50
And tell us if this fix your problem?
Fred

Post by Ioannis Marras
Hi all,
After the compilation of the training and validation function of my CNN I
ERROR (theano.gof.opt): EquilibriumOptimizer max'ed out by
'cut_gpua_constant_transfers'. You can safely raise the current threshold
of 5.000000 with the theano flag 'optdb.max_use_ratio'.
Although my CNN was running after this error I am wondering if I can
ignore this error or not....
What does this error means? (I found on the internet that this is
related to T.set_subtensor(). In my code there are repeated applications
of T.set_subtensor())
Many Thanks,
Ioannis
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an
For more options, visit https://groups.google.com/d/optout.

[Frédéric Bastien]

Ideally, we should make this work out of the box in Theano.

I don't have time now to investigate this case, but could you make an issue
on Theano with a function_dump:
http://www.deeplearning.net/software/theano/library/compile/function.html#theano.compile.function.function_dump

That way, when we have the time, we can look at it.

thanks

Fred

Post by Ioannis Marras
Hi Fred,
Many thanks for your help.
Using this Flag, this error disappeared.
Best,
Ioannis

Post by Frédéric Bastien
The function will run correctly after this. So you use it. But it could
optdb.max_use_ratio=50
And tell us if this fix your problem?
Fred

Post by Ioannis Marras
Hi all,
After the compilation of the training and validation function of my CNN
ERROR (theano.gof.opt): EquilibriumOptimizer max'ed out by
'cut_gpua_constant_transfers'. You can safely raise the current threshold
of 5.000000 with the theano flag 'optdb.max_use_ratio'.
Although my CNN was running after this error I am wondering if I can
ignore this error or not....
What does this error means? (I found on the internet that this is
related to T.set_subtensor(). In my code there are repeated
applications of T.set_subtensor())
Many Thanks,
Ioannis
--
---
You received this message because you are subscribed to the Google
Groups "theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send
For more options, visit https://groups.google.com/d/optout.

--

---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an
For more options, visit https://groups.google.com/d/optout.