Pytests fails - random seed error - mac os X 10.10.5

[nicofarr]

Installation seems to work all the way, up to :

py.test --pyargs modl

I've uploaded the logfile,

modl_tests_log.txt

Thanks

Nico

[arthurmensch]

Hi,

The package has never been tested on OS X (nor is it actually possible using any kind of free CI service) and I have no way of reproducing this. I may have an idea of the problem, I'll see what I can do.

[leodesigner]

I am getting the same errors under Ubuntu 14.04 Vagrant box.

modl/dict_fact.py:193: in fit
    self.prepare(n_samples=X.shape[0], X=dict_init)
modl/dict_fact.py:364: in prepare
    self.replacement, random_seed)
modl/utils/randomkit/sampler.pyx:33: in modl.utils.randomkit.sampler.Sampler.__cinit__ (modl/utils/randomkit/sampler.c:2072)
    self.random_state = RandomState(seed=random_seed)
modl/utils/randomkit/random_fast.pyx:55: in modl.utils.randomkit.random_fast.RandomState.__init__ (modl/utils/randomkit/random_fast.c:2528)
    self.seed(seed)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

>   raise ValueError("Wrong seed")
E   ValueError: Wrong seed

modl/utils/randomkit/random_fast.pyx:72: ValueError

Looks like we should merge two issues: https://github.com/arthurmensch/modl/issues/11:

[leodesigner]

I was able to fix ValueError by adding to modl/utils/randomkit/random_fast.pyx

        elif isinstance(seed, (int, long)):
            rk_seed(seed, self.internal_state)
        else:
            raise ValueError("Wrong seed")

Due we have a long integer init value for seed in sampler.pyx

Now I am getting only:

================================================================= FAILURES ==================================================================
____________________________________________________________ test_scale_patches _____________________________________________________________

    def test_scale_patches(): 
        patch_size = (8, 8, 3)
        n = 100
        shape = (n, ) + patch_size
        rs = check_random_state(0)
        X = rs.randn(*shape) 
        Y = scale_patches(X, with_mean=True, with_std=True, channel_wise=True) 
        assert_array_almost_equal(Y.mean(axis=(1, 2)), 0) 
>       assert_array_almost_equal(np.sum(Y ** 2, axis=(1, 2)), 1 / 3)
E       AssertionError: 
E       Arrays are not almost equal to 6 decimals
E       
E       (mismatch 100.0%)
E        x: array([[ 0.333333,  0.333333,  0.333333],
E              [ 0.333333,  0.333333,  0.333333],
E              [ 0.333333,  0.333333,  0.333333],...
E        y: array(0)

modl/preprocessing/tests/test_image.py:18: AssertionError
============================================== 1 failed, 51 passed, 4 skipped in 2.15 seconds ===============================================

Any ideas how to deal with this ? Or just ignore it ?

[arthurmensch]

Can you try it again on master ? Thanks !

[dohmatob]

@nicofarr are you using python 2xx or python 3xx. The code only works for py3 AFAIK.

[nicofarr]

Hi, Indeed, it only works with Python 3.x ! It now works here both on Os X 10.10.5 as well. Thx, Best Nico

…

On Fri, Jun 2, 2017 at 10:57 AM, DOHMATOB Elvis ***@***.***> wrote: @nicofarr <https://github.com/nicofarr> are you using python 2xx or python 3xx. The code only works for py3 AFAIK. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#10 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACgk8UgiGvMwf8iYGE6EIYxKb7EYeiGlks5r_86QgaJpZM4L3nLh> .