I'm trying to figure out if I can use the ENABLE_GLOBAL_EXPLAIN feature on a sparse column in BigQuery. This issue came up when created a classification model for text data. My feature column is a sparse vector that represents the term frequency only for the words that show up in each row. As a result, I don't have a separate column for each word, I have two columns, feature.word_index and feature.word_frequency.
The model trains and runs no problem, but when I try to query the model (something like this)
SELECT
*
FROM
ML.GLOBAL_EXPLAIN(MODEL text_dataset.logistic_reg_explain_classifier
)
Instead of getting the relevance for each term, I just get one single relevance score for "feature".
Does anyone know if there is a way to get the feature importances for each term, rather than for the column that holds the sparse array of terms and frequencies?