René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"difference in promotion between py2 and py3","articleBody":"Here's a weird one.  I have a histogram class that overloads += operator:\n\n``` python\n  bp::class_\u003chist_t\u003e (\"histogram\", bp::init\u003cdouble,double,double,bool\u003e (\n                                          (bp::arg (\"min\"),\n                                           bp::arg (\"max\"),\n                                           bp::arg (\"delta\"),\n                                           bp::arg (\"clip\")=true\n                                           )))\n    .def (bp::init\u003cnd::Array\u003cint64_t,1,1\u003e,double,double,double,bool\u003e())\n    .def (bp::self += hist_t::value_type())\n    .def (bp::self += nd::Array\u003cdouble,1,0\u003e())\n    .def (\"__call__\", \u0026hist_t::operator())\n    .def_readwrite (\"results\", \u0026hist_t::buckets)\n    .add_property (\"buckets\", \u0026get_buckets\u003chist_t,nd::Array\u003cdouble,1,1\u003e \u003e)\n    .def (\"mean\", \u0026hist_t::Mean)\n    .def (\"rms\", \u0026hist_t::RMS)\n    .def (\"cumulative\", \u0026hist_t::Cumulative\u003cnd::Array\u003cdouble,1,1\u003e \u003e)\n    .def (\"inv_cumulative\", \u0026hist_t::OneMinusCumulative\u003cnd::Array\u003cdouble,1,1\u003e \u003e)\n    .def_pickle(hist_pickle_suite\u003chist_t\u003e())\n\n    .def_readonly (\"min\", \u0026hist_t::min)\n    .def_readonly (\"max\", \u0026hist_t::max)\n    .def_readonly (\"delta\", \u0026hist_t::delta)\n    .add_property (\"rc\", \u0026getrc\u003chist_t\u003e) // just for info\n    ;\n```\n\nI call it:\n        remod_out_hist += np.sum (np.max (remod_out_cnts, axis=1))\n\nNow if we print out the type (np.sum (np.max ...)) on \npy2: `\u003ctype 'numpy.int64'\u003e`\n\nbut on py3: `\u003cclass 'numpy.int64'\u003e`\n\nWhen run on py2, the type numpy.int64 seems to be promoted to double, which matches the 1st self += (value_type should be double).\n\nBut when run on py3, the promotion fails:\nBoost.Python.ArgumentError: Python argument types in\n    histogram.**iadd**(histogram, numpy.int64)\ndid not match C++ signature:\n    **iadd**(boost::python::back_reference\u003cHistogram\u003clong, double, ndarray::Array\u003clong, 1, 1\u003e \u003e\u0026\u003e, ndarray::Array\u003cdouble, 1, 0\u003e)\n    **iadd**(boost::python::back_reference\u003cHistogram\u003clong, double, ndarray::Array\u003clong, 1, 1\u003e \u003e\u0026\u003e, double)\n\nIf I explicitly write\n        remod_out_hist += float(np.sum (np.max (remod_out_cnts, axis=1)))\n\nthen it runs on py3\n\nI'm wondering if the difference in type() on py2 and py3 might explain the different behavior.\n","author":{"url":"https://github.com/nbecker","@type":"Person","name":"nbecker"},"datePublished":"2015-06-11T18:02:00.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":1},"url":"https://github.com/41/ndarray/issues/41"}