Closed
Description
Running the following command on ppc64(be) hangs due to a near-infinite loop:
>>> np.random.hypergeometric([1,2,2], [2], [2])Appears to be related to new code in #13761, perhaps @WarrenWeckesser has an idea. See in particular frame #2 below. It appears a value is negative but the code expects it not to be.
gdb stack trace after breaking during hang:
#0 __log (x=-25811282) at ../math/w_log.c:27
#1 0x0ea14754 in loggam (x=-160848451)
at numpy/random/src/distributions/distributions.c:503
#2 0x0ea1232c in random_hypergeometric_hrua (bitgen_state=0xf76272c8, good=1,
bad=2, sample=268080756)
at numpy/random/src/legacy/legacy-distributions.c:266
#3 0x0ea12798 in random_hypergeometric_original (bitgen_state=0xf76272c8,
good=1, bad=2, sample=268080756)
at numpy/random/src/legacy/legacy-distributions.c:315
#4 0x0ea12844 in legacy_random_hypergeometric (bitgen_state=0xf76272c8,
good=1, bad=2, sample=8858015348)
at numpy/random/src/legacy/legacy-distributions.c:334
#5 0x0e96db10 in __pyx_f_5numpy_6random_6common_discrete_broadcast_iii (
__pyx_v_func=0xea127e8 <legacy_random_hypergeometric>,
__pyx_v_state=0xf76272c8, __pyx_v_size=0x104520e0 <_Py_NoneStruct>,
__pyx_v_lock=0xf7647140, __pyx_v_a_arr=0xf754c638,
__pyx_v_a_name=0xf7643fa0,
__pyx_v_a_constraint=__pyx_e_5numpy_6random_6common_CONS_NON_NEGATIVE,
__pyx_v_b_arr=0xf754c570, __pyx_v_b_name=0xf7643f80,
__pyx_v_b_constraint=__pyx_e_5numpy_6random_6common_CONS_NON_NEGATIVE,
__pyx_v_c_arr=0xf754c548, __pyx_v_c_name=0xf7643fe0,
__pyx_v_c_constraint=__pyx_e_5numpy_6random_6common_CONS_GTE_1)
at numpy/random/common.c:13924
---Type <return> to continue, or q <return> to quit---frame 1
#6 0x0e9e7c5c in __pyx_pf_5numpy_6random_6mtrand_11RandomState_96hypergeometric (__pyx_v_self=0xf76272b8, __pyx_v_ngood=0xf7551058, __pyx_v_nbad=0xf754ffa8,
__pyx_v_nsample=0xf754fe68, __pyx_v_size=0x104520e0 <_Py_NoneStruct>)
at numpy/random/mtrand.c:15834
#7 0x0e9e536c in __pyx_pw_5numpy_6random_6mtrand_11RandomState_97hypergeometric (__pyx_v_self=0xf76272b8, __pyx_args=0xf7535b48, __pyx_kwds=0x0)
at numpy/random/mtrand.c:15312
#8 0x1006ca2c in PyCFunction_Call ()
A little bit of gdb exploration so far:
(gdb) frame 1
#1 0x0ea14754 in loggam (x=-160848451)
at numpy/random/src/distributions/distributions.c:503
(gdb) p n
$2 = 160848458
(gdb) p x
$3 = -160848451
(gdb) frame 2
#2 0x0ea1232c in random_hypergeometric_hrua (bitgen_state=0xf76272c8, good=1,
bad=2, sample=268080756)
at numpy/random/src/legacy/legacy-distributions.c:266
266 d10 = (loggam(d9 + 1) + loggam(mingoodbad - d9 + 1) + loggam(m - d9 + 1) +
(gdb) l
261 d5 = 1.0 - d4;
262 d6 = m * d4 + 0.5;
263 d7 = sqrt((double)(popsize - m) * sample * d4 * d5 / (popsize - 1) + 0.5);
264 d8 = D1 * d7 + D2;
265 d9 = (RAND_INT_TYPE)floor((double)(m + 1) * (mingoodbad + 1) / (popsize + 2));
266 d10 = (loggam(d9 + 1) + loggam(mingoodbad - d9 + 1) + loggam(m - d9 + 1) +
267 loggam(maxgoodbad - m + d9 + 1));
268 d11 = MIN(MIN(m, mingoodbad) + 1.0, floor(d6 + 16 * d7));
269 /* 16 for 16-decimal-digit precision in D1 and D2 */
270
(gdb) p d9
$4 = -107232301
The failing line 266 in the last bit of code is from #13761, but I haven't read it well enough yet to know what it should do.