Now that mjolnir is now able to do queries and feature collection from 7.10.2 we are running into some new issues, exemplified by the following features collected from a single document:
'title_sum_raw_df': 0.0, 'title_min_raw_df': 3.4028235e+38, 'title_max_raw_df': 0.0, 'title_mean_raw_df': nan, 'title_stddev_raw_df': nan,
In this constructed example the query terms only matched the document body, but not the title. In 6.8.23 this returned all 0's, but in 7.10.2 we are getting Float.MAX_VALUE for min, and nan's for mean/stddev. Per the codebase the 7.10.2 version should still be returning 0's for all values there.
This will need to be fixed in the LTR plugin, fixing it only on the mjolnir side would introduce a variation (plausibly minimal) between the features used at training time and the features used at evaluation time. This almost certainly also effects the currently deployed models.