edflow.hooks.metric_hooks.tf_metric_hook module

Summary

Classes:

MetricHook

Applies a set of given metrics to the calculated data.

Reference

class edflow.hooks.metric_hooks.tf_metric_hook.MetricHook(metrics, save_root, consider_only_first=None)

Bases: edflow.hooks.hook.Hook

Applies a set of given metrics to the calculated data.

__init__(metrics, save_root, consider_only_first=None)

Parameters

• metrics (list) –

  List of MetricTuple``s of the form ``: input names, output names, metric, name).

  • input names are the keys corresponding to the feeds of

    interest, e.g. an original image.
  • output names are the keys corresponding to the values

    in the results dict.
  • metric is a Callable that accepts all inputs and

    outputs keys as keyword arguments

  • name is a

  If nested feeds or results are expected the names can be passed as “path” like 'key1_key2' returning dict[key1][key2].

• save_root (str) – Path to where the results are stored.

• consider_only_first (int) – Metric is only evaluated on the first consider_only_first examples.

before_epoch(epoch)

Called before each epoch.

Parameters

epoch (int) – Index of epoch that just started.

before_step(step, fetches, feeds, batch)

Called before each step. Can update any feeds and fetches.

Parameters

• step (int) – Current training step.

• fetches (list or dict) – Fetches for the next session.run call.

• feeds (dict) – Data used at this step.

• batch (list or dict) – All data available at this step.

after_step(step, results)

Called after each step.

Parameters

• step (int) – Current training step.

• last_results (list) – Results from last time this hook was called.

after_epoch(epoch)

Called after each epoch.

Parameters

epoch (int) – Index of epoch that just ended.