[0.20.2] On armhf, test_novelty_true_common_tests() leads to a segfault #13052

olebole · 2019-01-27T10:31:53Z

Splitting #13036: On armhf, test_novelty_true_common_tests() probably leads to a segfault. I will try to get a backtrace.
Commenting this out leads to the next segfault:

sklearn/tests/test_common.py::test_non_meta_estimators[BaggingRegressor-BaggingRegressor-check_regressors_train(readonly_memmap=True)] Bus error
E: pybuild pybuild:338: test: plugin distutils failed with: exit code=135: cd /<<BUILDDIR>>/scikit-learn-0.20.2+dfsg/.pybuild/cpython2_2.7/build; python2.7 -m pytest -m "not network" -v

and the next is

sklearn/utils/tests/test_estimator_checks.py::test_check_estimator_transformer_no_mixin PASSED [ 97%]
sklearn/utils/tests/test_estimator_checks.py::test_check_estimator_clones Bus error
E: pybuild pybuild:338: test: plugin distutils failed with: exit code=135: cd /<<BUILDDIR>>/scikit-learn-0.20.2+dfsg/.pybuild/cpython2_2.7/build; python2.7 -m pytest -m "not network" -v
dh_auto_test: pybuild --test -i python{version} -p 2.7 returned exit code 13

...

sklearn/utils/tests/test_estimator_checks.py::test_check_estimator_pairwise Bus error
E: pybuild pybuild:338: test: plugin distutils failed with: exit code=135: cd /<<BUILDDIR>>/scikit-learn-0.20.2+dfsg/.pybuild/cpython2_2.7/build; python2.7 -m pytest -m "not network" -v
dh_auto_test: pybuild --test -i python{version} -p 2.7 returned exit code 13

This list is reproducible; all other tests pass.

The text was updated successfully, but these errors were encountered:

olebole · 2019-01-29T13:30:33Z

Stacktrace for the test_non_meta_estimators (Program received signal SIGBUS, Bus error) is (on Python 2.7)

(gdb) bt
#0  __pyx_f_7sklearn_4tree_10_criterion_19RegressionCriterion_init (__pyx_v_self=0xb0db2e90, __pyx_v_y=0xb6fee88a, __pyx_v_y_stride=1, __pyx_v_sample_weight=0x20d3df8, 
    __pyx_v_weighted_n_samples=<optimized out>, __pyx_v_samples=0x2102a38, __pyx_v_start=0, __pyx_v_end=128) at sklearn/tree/_criterion.c:6785
#1  0xb351b18c in __pyx_f_7sklearn_4tree_9_splitter_8Splitter_node_reset (__pyx_v_self=0xb2a0ac90, __pyx_v_start=<optimized out>, __pyx_v_end=<optimized out>, 
    __pyx_v_weighted_n_node_samples=0xbeff9c90) at sklearn/tree/_splitter.c:4075
#2  0xb34ead38 in __pyx_f_7sklearn_4tree_5_tree_21DepthFirstTreeBuilder_build (__pyx_v_self=<optimized out>, __pyx_v_tree=0xb0ab6ad8, __pyx_v_X=0xb220f098, 
    __pyx_v_y=0xb220f200, __pyx_skip_dispatch=1, __pyx_optional_args=0xbeff9d7c) at sklearn/tree/_tree.c:5387
#3  0xb34e4b6a in __pyx_pf_7sklearn_4tree_5_tree_21DepthFirstTreeBuilder_2build (__pyx_v_X_idx_sorted=<optimized out>, __pyx_v_sample_weight=<optimized out>, 
    __pyx_v_y=<optimized out>, __pyx_v_X=<optimized out>, __pyx_v_tree=<optimized out>, __pyx_v_self=0xb0aa4780) at sklearn/tree/_tree.c:6034
#4  __pyx_pw_7sklearn_4tree_5_tree_21DepthFirstTreeBuilder_3build (__pyx_v_self=0xb0aa4780, __pyx_args=<optimized out>, __pyx_kwds=<optimized out>)
    at sklearn/tree/_tree.c:6005
#5  0x0046f19c in PyEval_EvalFrameEx ()
#6  0x0046d178 in PyEval_EvalCodeEx ()
#7  0x004733a0 in PyEval_EvalFrameEx ()
#8  0x0046d178 in PyEval_EvalCodeEx ()
#9  0x004733a0 in PyEval_EvalFrameEx ()
#10 0x0046d178 in PyEval_EvalCodeEx ()

Some variables (from a different session, so the pointer values do not match the stacktrace above):

(gdb) p *__pyx_v_self
$13 = {__pyx_base = {ob_refcnt = 2, 
    ob_type = 0xb34f87c4 <__pyx_type_7sklearn_4tree_10_criterion_MSE>, 
    __pyx_vtab = 0xb34f8cd0 <__pyx_vtable_7sklearn_4tree_10_criterion_MSE>, 
    y = 0xb6f9288a, y_stride = 1, sample_weight = 0x2c9f090, samples = 0x2c92aa0, 
    start = 0, pos = 0, end = 128, n_outputs = 1, n_samples = 200, 
    n_node_samples = 128, weighted_n_samples = 200, weighted_n_node_samples = 0, 
    weighted_n_left = 0, weighted_n_right = 0, sum_total = 0x2c38c88, 
    sum_left = 0x2535ed0, sum_right = 0x2c54488}, sq_sum_total = 0}
(gdb) p *__pyx_v_y
Cannot access memory at address 0xb6f9288a
(gdb) p *__pyx_v_sample_weight 
$17 = 1

Interestingly, __pyx_v_y is unaccessible here -- this is the first argument of the call in

scikit-learn/sklearn/tree/_splitter.pyx

Lines 208 to 214 in fdf2f38

    
           self.criterion.init(self.y, 
        
                               self.y_stride, 
        
                               self.sample_weight, 
        
                               self.weighted_n_samples, 
        
                               self.samples, 
        
                               start, 
        
                               end)

amueller · 2019-01-29T16:30:15Z

Is this python2.7? Is this reproducible under 3.6 or 3.7?

olebole · 2019-01-29T16:38:14Z

That is 2.7. I will check tomorrow (or so) how it behaves under 3.7.

jnothman · 2019-01-30T00:33:14Z

My cythonised version of _criterion.pyx doesn't have code on line 6785. Could you provide the pyx context @olebole?

olebole · 2019-01-30T12:25:26Z

The sigbus also happens under Python-3.7. I attach the built files _criterion.c, _splitter.c, and _tree.c.

jnothman · 2019-01-30T13:01:03Z

Does this happen when you run test_novelty_true_common_tests alone? I don't get how this arises from that test, which shouldn't use sklearn/tree at all.

olebole · 2019-01-30T13:22:23Z

That is the second one (test_non_meta_estimators[BaggingRegressor-BaggingRegressor-check_...]), sorry. I unfortunately didn't read my own issue title, but just the first example.

jnothman · 2019-01-30T20:58:03Z

Right. I only see now that you edited the description.

test_check_estimator_pairwise doesn't have any trees either, but could be related to the LOF issue?

olebole · 2019-01-31T07:48:10Z

If that was a question to me, then I didn't understant what you want ;-)

adrinjalali · 2020-04-20T16:09:52Z

clearing from the milestone.

adrinjalali · 2021-10-05T09:17:02Z

Closing this, happy to reopen if we can reproduce it.

rth added the Bug label Jan 27, 2019

rth added this to the 0.20.3 milestone Jan 27, 2019

jnothman modified the milestones: 0.20.3, 0.21.1 Apr 6, 2019

rth modified the milestones: 0.21.1, 0.21.2 May 18, 2019

jnothman modified the milestones: 0.21.2, 0.21.3 May 25, 2019

jnothman modified the milestones: 0.21.3, 0.23 Oct 31, 2019

cmarmo mentioned this issue Mar 29, 2020

2 test failures on Debian stable (stretch) amd64 #12548

Closed

adrinjalali added the module:tree label Apr 20, 2020

adrinjalali modified the milestones: 0.23, 0.24 Apr 20, 2020

adrinjalali closed this as completed Oct 5, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[0.20.2] On armhf, test_novelty_true_common_tests() leads to a segfault #13052

[0.20.2] On armhf, test_novelty_true_common_tests() leads to a segfault #13052

olebole commented Jan 27, 2019 •

edited

Loading

olebole commented Jan 29, 2019 •

edited

Loading

amueller commented Jan 29, 2019

olebole commented Jan 29, 2019

jnothman commented Jan 30, 2019

olebole commented Jan 30, 2019

jnothman commented Jan 30, 2019 via email

olebole commented Jan 30, 2019

jnothman commented Jan 30, 2019

olebole commented Jan 31, 2019

adrinjalali commented Apr 20, 2020

adrinjalali commented Oct 5, 2021

[0.20.2] On armhf, test_novelty_true_common_tests() leads to a segfault #13052

[0.20.2] On armhf, test_novelty_true_common_tests() leads to a segfault #13052

Comments

olebole commented Jan 27, 2019 • edited Loading

olebole commented Jan 29, 2019 • edited Loading

amueller commented Jan 29, 2019

olebole commented Jan 29, 2019

jnothman commented Jan 30, 2019

olebole commented Jan 30, 2019

jnothman commented Jan 30, 2019 via email

olebole commented Jan 30, 2019

jnothman commented Jan 30, 2019

olebole commented Jan 31, 2019

adrinjalali commented Apr 20, 2020

adrinjalali commented Oct 5, 2021

olebole commented Jan 27, 2019 •

edited

Loading

olebole commented Jan 29, 2019 •

edited

Loading