def evaluate_post_training(self, edp: EvaluationDataPage) -> CpeDetails:
cpe_details = CpeDetails()
cpe_details.reward_estimates = self.score_cpe("Reward", edp)
if (
self.metrics_to_score is not None
and edp.logged_metrics is not None
and self.action_names is not None
):
for i, metric in enumerate(self.metrics_to_score):
logger.info(
"--------- Running CPE on metric: {} ---------".format(metric)
)
metric_reward_edp = edp.set_metric_as_reward(i, len(self.action_names))
cpe_details.metric_estimates[metric] = self.score_cpe(
metric, metric_reward_edp
)
if self.action_names is not None:
if edp.optimal_q_values is not None:
value_means = edp.optimal_q_values.mean(dim=0)
cpe_details.q_value_means = {
action: float(value_means[i])
for i, action in enumerate(self.action_names)
}
value_stds = edp.optimal_q_values.std(dim=0)
cpe_details.q_value_stds = {
action: float(value_stds[i])
for i, action in enumerate(self.action_names)
}
if edp.eval_action_idxs is not None:
cpe_details.action_distribution = {
# pyre-fixme[6]: Expected `Union[_SupportsIndex, bytearray,
# bytes, str, typing.SupportsFloat]` for 1st param but got
# `ByteTensor`.
action: float((edp.eval_action_idxs == i).sum())
/ edp.eval_action_idxs.shape[0]
for i, action in enumerate(self.action_names)
}
# Compute MC Loss on Aggregate Reward
cpe_details.mc_loss = float(
F.mse_loss(edp.logged_values, edp.model_values_for_logged_action)
)
# pyre-fixme[16]: `Evaluator` has no attribute `notify_observers`.
self.notify_observers(cpe_details=cpe_details)
return cpe_details
def evaluate_post_training(self, edp: EvaluationDataPage) -> CpeDetails:
cpe_details = CpeDetails()
cpe_details.reward_estimates = self.score_cpe("Reward", edp)
if (
self.metrics_to_score is not None
and edp.logged_metrics is not None
and self.action_names is not None
):
for i, metric in enumerate(self.metrics_to_score):
logger.info(
"--------- Running CPE on metric: {} ---------".format(metric)
)
metric_reward_edp = edp.set_metric_as_reward(i, len(self.action_names))
cpe_details.metric_estimates[metric] = self.score_cpe(
metric, metric_reward_edp
)
if self.action_names is not None:
if edp.optimal_q_values is not None:
value_means = edp.optimal_q_values.mean(dim=0)
cpe_details.q_value_means = {
action: float(value_means[i])
for i, action in enumerate(self.action_names)
}
# pyre-ignore [16]: `Optional` has no attribute `std`
value_stds = edp.optimal_q_values.std(dim=0)
cpe_details.q_value_stds = {
action: float(value_stds[i])
for i, action in enumerate(self.action_names)
}
if edp.eval_action_idxs is not None:
cpe_details.action_distribution = {
# pyre-ignore [16]: `bool` has no attribute `sum`
action: float((edp.eval_action_idxs == i).sum())
# pyre-ignore [16]: `Optional` has no attribute `shape`
/ edp.eval_action_idxs.shape[0]
for i, action in enumerate(self.action_names)
}
# pyre-fixme[16]: `Evaluator` has no attribute `notify_observers`.
self.notify_observers(cpe_details=cpe_details)
return cpe_details
def evaluate_post_training(self, edp: EvaluationDataPage) -> CpeDetails:
cpe_details = CpeDetails()
cpe_details.reward_estimates = self.score_cpe("Reward", edp)
if (self.metrics_to_score is not None
and edp.logged_metrics is not None
and self.action_names is not None):
for i, metric in enumerate(self.metrics_to_score):
logger.info(
"--------- Running CPE on metric: {} ---------".format(
metric))
metric_reward_edp = edp.set_metric_as_reward(
i, len(self.action_names))
cpe_details.metric_estimates[metric] = self.score_cpe(
metric, metric_reward_edp)
if self.action_names is not None:
if edp.optimal_q_values is not None:
value_means = edp.optimal_q_values.mean(dim=0)
cpe_details.q_value_means = {
action: float(value_means[i])
for i, action in enumerate(self.action_names)
}
value_stds = edp.optimal_q_values.std(dim=0) # type: ignore
cpe_details.q_value_stds = {
action: float(value_stds[i])
for i, action in enumerate(self.action_names)
}
if edp.eval_action_idxs is not None:
cpe_details.action_distribution = {
action: float(
(edp.eval_action_idxs == i).sum()) # type: ignore
/ edp.eval_action_idxs.shape[0]
for i, action in enumerate(self.action_names)
}
# Compute MC Loss on Aggregate Reward
cpe_details.mc_loss = float(
F.mse_loss(edp.logged_values, edp.model_values_for_logged_action))
self.notify_observers(cpe_details=cpe_details) # type: ignore
return cpe_details
