Function align_predictions

examples/token-classification/run_ner.py:192–206

(predictions: np.ndarray, label_ids: np.ndarray)

    )

    def align_predictions(predictions: np.ndarray, label_ids: np.ndarray) -> Tuple[List[int], List[int]]:
        preds = np.argmax(predictions, axis=2)

        batch_size, seq_len = preds.shape

        out_label_list = [[] for _ in range(batch_size)]
        preds_list = [[] for _ in range(batch_size)]

        for i in range(batch_size):
            for j in range(seq_len):
                if label_ids[i, j] != nn.CrossEntropyLoss().ignore_index:
                    out_label_list[i].append(label_map[label_ids[i][j]])
                    preds_list[i].append(label_map[preds[i][j]])

        return preds_list, out_label_list

    def compute_metrics(p: EvalPrediction) -> Dict:
        preds_list, out_label_list = align_predictions(p.predictions, p.label_ids)