SHAP C++ Machine Learning Text Classification
Budget: $1,500 – $3,000 USD
Objective: Write equivalent of the SHAP python algorithm in C++. (THIS NEEDS TO BE DONE FAST, SO IF YOU CANNOT START RIGHT AWAY AND PUT A LOT OF FOCUS THEN YOU NEED NOT APPLY)
Method: You will have to create your own dataset (download text from cloud database of your choice) and model in C++. This project is mostly about the SHAP algorithm but needs to be able to be tested locally so the other components need to be produced at least to test it. So I don't care if you use pytorch c++ or what model you use and the details behind that. Whatever gets the job done fastest.
Deliverable:I need the functions well documented to run. Probably need something LIKE the following:
shap.h: all of the information to instantiate and store, etc...
train_model.cpp: calls load data function, instantiates c++bert/wordembedding transformers..., trains ANN
explain_cpp: instantiates SHAP, creates local predication on a string of text
NOTE: ADD NOTE THE MEMORY ACCESSING AND ALLOCATION BECAUSE WE WILL NEED TO PARALLELIZE BECAUSE WE HAVE CUSTOM HARDWARE TO PARALLELIZE THOUSANDS OF THREADS SO NOTING THESE POSSIBILITIES WILL HELP BUT IS NOT PART OF THE SCOPE.)
Method: You will have to create your own dataset (download text from cloud database of your choice) and model in C++. This project is mostly about the SHAP algorithm but needs to be able to be tested locally so the other components need to be produced at least to test it. So I don't care if you use pytorch c++ or what model you use and the details behind that. Whatever gets the job done fastest.
Deliverable:I need the functions well documented to run. Probably need something LIKE the following:
shap.h: all of the information to instantiate and store, etc...
train_model.cpp: calls load data function, instantiates c++bert/wordembedding transformers..., trains ANN
explain_cpp: instantiates SHAP, creates local predication on a string of text
NOTE: ADD NOTE THE MEMORY ACCESSING AND ALLOCATION BECAUSE WE WILL NEED TO PARALLELIZE BECAUSE WE HAVE CUSTOM HARDWARE TO PARALLELIZE THOUSANDS OF THREADS SO NOTING THESE POSSIBILITIES WILL HELP BUT IS NOT PART OF THE SCOPE.)