-
-
Notifications
You must be signed in to change notification settings - Fork 9.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
cannot pass theano Unit Test by using the numpy code 12/26/2016 on cpu #8420
Comments
Looks like Theano hasn't kept up with |
I don't really like that we are suddenly turning this into an error, it breaks a lot of stuff. |
@carletes Thanks very much for your quick feedback.
|
@juliantaylor I only changed things that were warned about in NumPy 1.9 or earlier, i.e., at least two years ago. @KevinYuk Yes, there appear to be other errors, arctanh has changed, it is now black listed for more glibc versions, and there may be errors that are genuinely on the numpy side. I haven't looked too closely yet. BTW, best to avoid lines starting with |
@KevinYuk I'd be much obliged if you could run tests against the |
@KevinYuk Relaxed stride checking also seems to be a problem. BTW, these changes in master will be in the 1.13 release, so probably 4-6 months off. |
@charris Let's try maintenance/1.12.x branch. One second. |
@charris Please check it by using maintenance/1.12.x branch, still report error and fail. It seems that 1.11x works fine. So maybe something wrong happened from 1.11.x to 1.12.x changing. log:
|
Thanks for running the tests! It looks like 1.12 will be a problem for theano. The indexing with a float should be easy to fix, and I suspect the arctanh test just requires a change in precision, not sure about the other failures. Do you know what version of glibc you have? |
@charris Please check below: |
Yep, glibc > 2.17 is now needed to avoid the numpy fallback functions. It was previously accepted as the earliest good version, but some precision problems still turned up. See
|
@charris Thanks for your comments. Do you mean I should use glibc < 2.17 to avoid the related UT errors? Or is this a glibc 2.17 bug? |
Hmm, I'm wrong. The functions blacklisted are the complex versions, which don't seem to be tested. So I'm not sure what is going on. However
That leaves the arctanh errors and I don't know what is going on there. |
Thanks for all that investigation and help. I did an issue on Theano
related to that:
Theano/Theano#5396
We are preparing a release candidate. I'll try to have this done for it.
…On Tue, Jan 3, 2017 at 8:03 PM, Charles Harris ***@***.***> wrote:
Hmm, I'm wrong. The functions blacklisted are the complex versions, which
don't seem to be tested. So I'm not sure what is going on. However
- TypeError: 'float' object cannot be interpreted as an index errors
are from using floats to index arrays, it can probably be fixed in the test
by making the dimensions in out_shp integer.
- TypeError: Cannot cast array data from dtype('float32') to
dtype('int64') according to the rule 'safe' can be fixed by making the
cast explicit, casting="unsafe".
- TypeError: invalid strides looks to be because of relaxed stride
checking.
That leaves the arctanh errors and I don't know what is going on there.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#8420 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AALC-82Laf8aRqLNrcuBFde0AnmvicMBks5rOu_kgaJpZM4LVf-5>
.
|
We're working on fixing the Theano issue, but can't reproduce "TypeError: 'float' object cannot be interpreted as an index" and the arctanh errors. Could it be old glibc issues? |
It seems development of Theno has stopped. Can we close this? |
Closing. Please reopen if needed. |
Important notes(can pass theano UT by using numpy release version):
CAN PASS theano Unit Test by using numpy release versoin numpy-1.11.2rc1.
For save your time, you guys can also do quick verification to check the theano UT ERROR and FAIL by below methods.(Surely, it's only include partial error report):
Quick verification method:
(Please ingore the two CAReduce errors below. It's caused by offical theano:
FAIL: test_c (theano.tensor.tests.test_elemwise.test_CAReduce)
FAIL: test_perform (theano.tensor.tests.test_elemwise.test_CAReduce)
)
The text was updated successfully, but these errors were encountered: