[GSOC][TMVA][SOFIE] Cast ONNX Operator implemented with the corresponding unit tests #11033
Conversation
Added the functionality and support of int input type in Cast ONNX Operator
Neel-Shah-29
requested review from
bellenot,
lmoneta and
sitongan
as code owners
|
Can one of the admins verify this patch? |
|
@phsft-bot build just on ROOT-ubuntu2204/default, ROOT-ubuntu2004/default, ROOT-ubuntu18.04/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu18.04/default. Errors:
|
|
Build failed on ROOT-ubuntu2004/default. Errors:
|
|
Build failed on ROOT-ubuntu2204/default. Errors:
|
|
@phsft-bot build just on ROOT-ubuntu2204/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu2204/default. Failing tests:
|
|
@phsft-bot build just on ROOT-ubuntu2204/default, ROOT-ubuntu2004/default, ROOT-ubuntu18.04/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu2204/default. Errors:
|
|
Build failed on ROOT-ubuntu18.04/default. Errors:
|
|
Build failed on ROOT-ubuntu2004/default. Errors:
|
There was a problem hiding this comment.
@Neel-Shah-29 , please include the support for the other type of casting.
The generate code has also some issues, from the test one has:
//Code generated automatically by TMVA for Inference of Model file [Cast.onnx] at [Mon Aug 1 16:00:41 2022]
#ifndef TMVA_SOFIE_CAST
#define TMVA_SOFIE_CAST
#include<vector>
#include "TMVA/SOFIE_common.hxx"
#include <fstream>
namespace TMVA_SOFIE_Cast{
struct Session {
Session(std::string filename ="") {
if (filename.empty()) filename = "Cast.dat";
}
infer){
//------ CAST
for (int id = 0; id < 6 ; id++){
tensor_1[id] = (float)(tensor_onnxCast0[id]);
}
std::vector<float> ret (tensor_1, tensor_1 + 6);
return ret;
}
};
} //TMVA_SOFIE_Cast
the infer function is not correctly defined. This is happening in RModel, because it sees that there is no output tensor. I think in this case a void infer function should be generated, I will fix this.
However, the input model should have an output tensor. This is maybe a possible problem with the generated test model. Can you please check this too
…Parser for supporting different input types
…operator and added the support to RModel::Generate method also
|
@phsft-bot build just on ROOT-ubuntu2204/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu2204/default. Warnings:
Failing tests:
|
This fixes the new Cast operator. Several changes are needed for Cast since the input tensor can be of a type different than float Also a fix for parsing correctly the attribute of Cast is applied
|
@phsft-bot build just on ROOT-ubuntu2204/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu2204/default. Warnings:
Failing tests:
|
|
@phsft-bot build just on ROOT-ubuntu2204/default with flags -Dtmva-sofie=On |
|
Starting build on |
|
Build failed on ROOT-ubuntu2204/default. Warnings:
Failing tests:
|
This Pull request: Adds the Cast ONNX Operator with the corresponding Unit tests to validate the written code.
Checklist: