A single step of evaluation.
()
| 145 | captured_scaffold_fn = _CapturedObject() |
| 146 | |
| 147 | def eval_step(): |
| 148 | """A single step of evaluation.""" |
| 149 | estimator_spec = self._call_model_fn(features, labels, |
| 150 | model_fn_lib.ModeKeys.EVAL, params) |
| 151 | |
| 152 | try: |
| 153 | captured_scaffold_fn.capture(estimator_spec.scaffold_fn) |
| 154 | except AttributeError: |
| 155 | captured_scaffold_fn.capture(None) |
| 156 | |
| 157 | eval_metric_fn = None |
| 158 | eval_metric_fn_tensors = [] |
| 159 | try: |
| 160 | if estimator_spec.eval_metrics: |
| 161 | (eval_metric_fn, eval_metric_fn_tensors) = estimator_spec.eval_metrics |
| 162 | except AttributeError: |
| 163 | pass |
| 164 | |
| 165 | # If a dictionary is provided, we need to convert it into a list sorted |
| 166 | # according to order of eval_metric_fn positional arguments. |
| 167 | if isinstance(eval_metric_fn_tensors, dict): |
| 168 | eval_metric_fn_args = function_utils.fn_args(eval_metric_fn) |
| 169 | eval_metric_fn_tensors = [ |
| 170 | eval_metric_fn_tensors[i] for i in eval_metric_fn_args |
| 171 | ] |
| 172 | |
| 173 | captured_eval_metric_fn.capture(eval_metric_fn) |
| 174 | |
| 175 | return tuple([estimator_spec.loss] + eval_metric_fn_tensors) |
| 176 | |
| 177 | return eval_step, captured_eval_metric_fn, captured_scaffold_fn |
| 178 |
nothing calls this directly
no test coverage detected