def dump_parsed_inputs()

in tzrec/datasets/data_parser.py [0:0]

• 31 lines of code
• 10 McCabe index (conditional complexity)

    def dump_parsed_inputs(self, input_data: Dict[str, torch.Tensor]) -> pa.Array:
        """Dump parsed inputs for debug."""
        feature_rows = defaultdict(dict)
        for f in self._features:
            if f.is_sparse:
                lengths = input_data[f"{f.name}.lengths"]
                values = input_data[f"{f.name}.values"].cpu().numpy()
                cnt = 0
                # pyre-ignore [16]
                sep = f.sequence_delim if f.is_sequence else ","
                for i, ll in enumerate(lengths):
                    cur_v = values[cnt : cnt + ll]
                    cnt += ll
                    feature_rows[i][f.name] = sep.join(cur_v.astype(str))
            else:
                if f.is_sequence:
                    lengths = input_data[f"{f.name}.lengths"]
                    values = input_data[f"{f.name}.values"].cpu().numpy()
                    cnt = 0
                    for i, ll in enumerate(lengths):
                        cur_v = values[cnt : cnt + ll]
                        cnt += ll
                        feature_rows[i][f.name] = f.sequence_delim.join(
                            map(",".join, cur_v.astype(str))
                        )
                else:
                    values = input_data[f"{f.name}.values"].cpu().numpy()
                    for i, cur_v in enumerate(values):
                        feature_rows[i][f.name] = ",".join(cur_v.astype(str))

        result = []
        for i in range(len(feature_rows)):
            result.append(" | ".join([f"{k}:{v}" for k, v in feature_rows[i].items()]))
        return pa.array(result)