afe.apis.statistic

Analysis of statistics on tensors.

Attributes

Metric

Functions

`equality`(→ float)	Equality as a distance metric.
`mean_float`(→ float)	Mean value of difference between input values and ground truth values.
`threshold_test_counter`(...)	Create a Statistic over a stream of (x, y) pairs
`tensor_set_statistics`(...)	Apply an independent Statistic to each tensor in a stream of pairs of
`mean`(...)	Create a statistic that takes input pairs (i, g)
`mean_text`(...)	Create a statistic that takes input pairs (i, g)

Module Contents

afe.apis.statistic.Metric[source]

afe.apis.statistic.equality(x: afe.apis.compilation_job_base.Tensor, y: afe.apis.compilation_job_base.Tensor) → float[source]

Equality as a distance metric.

Return 0 if the tensors are equal, 1 otherwise.

afe.apis.statistic.mean_float(x: afe.apis.compilation_job_base.Tensor, y: afe.apis.compilation_job_base.Tensor) → float[source]: Mean value of difference between input values and ground truth values.

afe.apis.statistic.threshold_test_counter(metric: Metric, threshold: float) → afe.driver.statistic.Statistic[Tuple[afe.apis.compilation_job_base.Tensor, afe.apis.compilation_job_base.Tensor], str][source]

Create a Statistic over a stream of (x, y) pairs that counts the number of times metric(x, y) < threshold is satisfied.

Parameters:

metric – Distance metric
threshold – Threshold to compare against

Returns:

Statistic that captures the data

afe.apis.statistic.tensor_set_statistics(statistics: List[afe.driver.statistic.Statistic[Tuple[Any, Any], str]]) → afe.driver.statistic.Statistic[Tuple[List[Any], List[Any]], str][source]

Apply an independent Statistic to each tensor in a stream of pairs of fixed-length lists.

This is intended for evaluating models that have multiple outputs and a ground truth value corresponding to each output. Each Statistic would be applied to one of the outputs and ground truth values.

Parameters:: statistics – Statistic to apply to each pair of values
Returns:: Composed statistic that applies the statistics to list items

afe.apis.statistic.mean(metric: Metric) → afe.driver.statistic.Statistic[Tuple[List[afe.apis.compilation_job_base.Tensor], afe.apis.compilation_job_base.Tensor], float][source]

Create a statistic that takes input pairs (i, g) and computes the arithmetic mean of metric(i, g) over all given inputs.

Parameters:: metric – Mean metric.
Returns:: Statistic that captures the data

afe.apis.statistic.mean_text(metric: Metric) → afe.driver.statistic.Statistic[Tuple[List[afe.apis.compilation_job_base.Tensor], afe.apis.compilation_job_base.Tensor], str][source]

Create a statistic that takes input pairs (i, g) and computes the arithmetic mean of metric(i, g) over all given inputs and formats the results as text message.

Parameters:: metric – Mean metric.
Returns:: Statistic that captures the data