train_final_model: Compare hyperparameter performance and train a final model

In mdneuzerling/NLPRMetaflow: What the Package Does (One Line, Title Case)

Description

Compare hyperparameter performance and train a final model

Usage

train_final_model(self, inputs)

Arguments

self	Metaflow state variable
inputs	Inputs from the previous split Metaflow steps

Details

This step is join in the flow, and so must first reduce the various hyperparameter evaluations produced in the previous steps and merge the other variables.

It appears as though Metaflow struggles with nested tibbles, likely due to a restriction with reticulate/Python. This prevents us from using the tuning results from tune directly, so we have to use a custom select_best_hyperparameters function.

mdneuzerling/NLPRMetaflow documentation built on Oct. 15, 2021, 10:14 a.m.