Each model in the cfb_model_artifacts bundle ships a <model>.card.json
carrying the ordered features array it was trained with and, where the
model consumes one, an era_contract. Reading that contract lets a caller's
frame be validated against the artifact itself rather than a feature list
restated in this package – the duplication that produced
cfbfastR-cfb-data#70, where this package kept a private 2017 era cut the
trainer never used and 2018-2020 scored an era off the models trained with
them.
cfb_model_card(): the parsed card.cfb_card_features(): the model's feature names, in trained order.cfb_card_era_contract(): the era encoding, orNULL.cfb_card_reset_cache(): drop the in-session card cache.
Usage
cfb_model_card(model)
cfb_card_features(model)
cfb_card_era_contract(model)
cfb_card_reset_cache()Value
cfb_model_card() - A list with the card's fields, including:
| col_name | types | description |
| model_type | character | Model family, e.g. "ep", "fg", "xpass". |
| label | character | Training label the model was fit against. |
| features | character | Ordered feature names the booster was trained with. |
| era_contract | list | Era encoding and cutpoints, or NULL when the model uses no era. |
cfb_card_features() - A character vector of feature names in the
order the booster was trained with. Order is load-bearing: xgboost aligns a
DMatrix by position, so a sorted or re-derived list scores against the
wrong columns without ever erroring.
cfb_card_era_contract() - A list with encoding ("ordinal" or
"one_hot"), columns and cuts, or NULL when the model consumes no
era feature. NULL is correct for ep_model, wp_naive, wp_spread and
cfb_cp_model.
cfb_card_reset_cache() - Invisibly NULL, called for its effect.
See also
Other CFB Model Calculators:
calculate_cfb_models
Examples
# \donttest{
try(cfb_card_features("ep_model"))
#> [1] "TimeSecsRem" "yards_to_goal" "distance"
#> [4] "down_1" "down_2" "down_3"
#> [7] "down_4" "pos_score_diff_start"
# }