[
https://issues.apache.org/jira/browse/MADLIB-835?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Frank McQuillan updated MADLIB-835:
-----------------------------------
Assignee: (was: Nandish Jayaram)
> Improved error message for Elastic Net predict()
> ------------------------------------------------
>
> Key: MADLIB-835
> URL: https://issues.apache.org/jira/browse/MADLIB-835
> Project: Apache MADlib
> Issue Type: Bug
> Reporter: Srivatsan
> Priority: Minor
> Labels: gsoc2016, starter
>
> When we pass the selected coefficients to elastic net's "predict()" function,
> it throws as ugly error message which is not indicative of the real error.
> {noformat}
> NOTICE: table "regression_elastic_net_predictions" does not exist, skipping
> ERROR: Function "madlib.elastic_net_gaussian_predict(double
> precision[],double precision,double precision[])": Internal error. Eigen
> assertion failed (index >= 0 && index < size()) in function typename
> Eigen::internal::conditional<(bool)((Eigen::internal::traits<T>::Flags &
> Eigen::LvalueBit)), const typename Eigen::internal::traits<T>::Scalar&,
> typename Eigen::internal::conditional<Eigen::internal::is_arithmetic<typename
> Eigen::internal::traits<T>::Scalar>::value, typename
> Eigen::internal::traits<T>::Scalar, const typename
> Eigen::internal::traits<T>::Scalar>::type>::type
> Eigen::DenseCoeffsBase<Derived, 0>::operator()(typename
> Eigen::internal::traits<T>::Index) const [with Derived = Eigen::Map<const
> Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>,
> 0, Eigen::Stride<0, 0> >] at
> /data/home/gpdbchina/pulse2-data/agents/agent1/work/MADlib%20Build%20%28Release%29/rhel5_x86_64/build/third_party/src/EP_eigen/Eigen/src/Core/DenseCoeffsBase.h:173
> (UDF_impl.hpp:208) (seg40 sdw6:1025 pid=21706) (cdbdisp.c:1477)
> DETAIL: PL/pgSQL function "elastic_net_predict" line 8 at return
> {noformat}
> The error is in passing "coef_nonzero" as input instead of "coef_all" from
> the model.
> {noformat}
> select test.person_uid,
> test.gpa as actual_gpa,
> madlib.elastic_net_predict(
> 'gaussian',
> mdl.coef_nonzero,
> mdl.intercept,
> feat_vect_normalized
> ) as predicted_gpa
> from modeling.regression_test test,
> modeling.regression_elastic_net_mdl mdl
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
