Returns a new config with default settings applied.
tfma.utils.update_eval_config_with_defaults(
eval_config: tfma.EvalConfig
,
maybe_add_baseline: Optional[bool] = None,
maybe_remove_baseline: Optional[bool] = None,
has_baseline: Optional[bool] = False,
rubber_stamp: Optional[bool] = False
) -> tfma.EvalConfig
a) Add or remove a model_spec according to "has_baseline".
b) Fix the model names (model_spec.name) to tfma.CANDIDATE_KEY and
tfma.BASELINE_KEY.
c) Update the metrics_specs with the fixed model name.
Args |
eval_config
|
Original eval config.
|
maybe_add_baseline
|
DEPRECATED. True to add a baseline ModelSpec to the
config as a copy of the candidate ModelSpec that should already be
present. This is only applied if a single ModelSpec already exists in the
config and that spec doesn't have a name associated with it. When applied
the model specs will use the names tfma.CANDIDATE_KEY and
tfma.BASELINE_KEY. Only one of maybe_add_baseline or maybe_remove_baseline
should be used.
|
maybe_remove_baseline
|
DEPRECATED. True to remove a baseline ModelSpec from
the config if it already exists. Removal of the baseline also removes any
change thresholds. Only one of maybe_add_baseline or maybe_remove_baseline
should be used.
|
has_baseline
|
True to add a baseline ModelSpec to the config as a copy of
the candidate ModelSpec that should already be present. This is only
applied if a single ModelSpec already exists in the config and that spec
doesn't have a name associated with it. When applied the model specs will
use the names tfma.CANDIDATE_KEY and tfma.BASELINE_KEY. False to remove a
baseline ModelSpec from the config if it already exists. Removal of the
baseline also removes any change thresholds. Only one of has_baseline or
maybe_remove_baseline should be used.
|
rubber_stamp
|
True if this model is being rubber stamped. When a model is
rubber stamped diff thresholds will be ignored if an associated baseline
model is not passed.
|
Raises |
RuntimeError
|
on missing baseline model for non-rubberstamp cases.
|