How to use the tiledb.dataframe_.ColumnInfo function in tiledb

def dtype_from_column(col):
    import pandas as pd

    col_dtype = col.dtype
    # TODO add more basic types here
    if col_dtype in (np.int32, np.int64, np.uint32, np.uint64, np.float, np.double,
                     np.uint8):
        return ColumnInfo(col_dtype)

    # TODO this seems kind of brittle
    if col_dtype.base == np.dtype('M8[ns]'):
        if col_dtype == np.dtype('datetime64[ns]'):
            return ColumnInfo(col_dtype)
        elif hasattr(col_dtype, 'tz'):
            raise ValueError("datetime with tz not yet supported")
        else:
            raise ValueError("unsupported datetime subtype ({})".format(type(col_dtype)))

    # Pandas 1.0 has StringDtype extension type
    if col_dtype.name == 'string':
        return ColumnInfo(unicode_dtype)

    if col_dtype == 'bool':
        return ColumnInfo(np.uint8, repr=np.dtype('bool'))

if col_dtype in (np.int32, np.int64, np.uint32, np.uint64, np.float, np.double,
                     np.uint8):
        return ColumnInfo(col_dtype)

    # TODO this seems kind of brittle
    if col_dtype.base == np.dtype('M8[ns]'):
        if col_dtype == np.dtype('datetime64[ns]'):
            return ColumnInfo(col_dtype)
        elif hasattr(col_dtype, 'tz'):
            raise ValueError("datetime with tz not yet supported")
        else:
            raise ValueError("unsupported datetime subtype ({})".format(type(col_dtype)))

    # Pandas 1.0 has StringDtype extension type
    if col_dtype.name == 'string':
        return ColumnInfo(unicode_dtype)

    if col_dtype == 'bool':
        return ColumnInfo(np.uint8, repr=np.dtype('bool'))

    if col_dtype == np.dtype("O"):
        # Note: this does a full scan of the column... not sure what else to do here
        #       because Pandas allows mixed string column types (and actually has
        #       problems w/ allowing non-string types in object columns)
        inferred_dtype = pd.api.types.infer_dtype(col)

        if inferred_dtype == 'bytes':
            return ColumnInfo(np.bytes_)

        elif inferred_dtype == 'string':
            # TODO we need to make sure this is actually convertible
            return ColumnInfo(unicode_dtype)

# TODO this seems kind of brittle
    if col_dtype.base == np.dtype('M8[ns]'):
        if col_dtype == np.dtype('datetime64[ns]'):
            return ColumnInfo(col_dtype)
        elif hasattr(col_dtype, 'tz'):
            raise ValueError("datetime with tz not yet supported")
        else:
            raise ValueError("unsupported datetime subtype ({})".format(type(col_dtype)))

    # Pandas 1.0 has StringDtype extension type
    if col_dtype.name == 'string':
        return ColumnInfo(unicode_dtype)

    if col_dtype == 'bool':
        return ColumnInfo(np.uint8, repr=np.dtype('bool'))

    if col_dtype == np.dtype("O"):
        # Note: this does a full scan of the column... not sure what else to do here
        #       because Pandas allows mixed string column types (and actually has
        #       problems w/ allowing non-string types in object columns)
        inferred_dtype = pd.api.types.infer_dtype(col)

        if inferred_dtype == 'bytes':
            return ColumnInfo(np.bytes_)

        elif inferred_dtype == 'string':
            # TODO we need to make sure this is actually convertible
            return ColumnInfo(unicode_dtype)

        elif inferred_dtype == 'mixed':
            raise ValueError(

def dtype_from_column(col):
    import pandas as pd

    col_dtype = col.dtype
    # TODO add more basic types here
    if col_dtype in (np.int32, np.int64, np.uint32, np.uint64, np.float, np.double,
                     np.uint8):
        return ColumnInfo(col_dtype)

    # TODO this seems kind of brittle
    if col_dtype.base == np.dtype('M8[ns]'):
        if col_dtype == np.dtype('datetime64[ns]'):
            return ColumnInfo(col_dtype)
        elif hasattr(col_dtype, 'tz'):
            raise ValueError("datetime with tz not yet supported")
        else:
            raise ValueError("unsupported datetime subtype ({})".format(type(col_dtype)))

    # Pandas 1.0 has StringDtype extension type
    if col_dtype.name == 'string':
        return ColumnInfo(unicode_dtype)

    if col_dtype == 'bool':
        return ColumnInfo(np.uint8, repr=np.dtype('bool'))

    if col_dtype == np.dtype("O"):
        # Note: this does a full scan of the column... not sure what else to do here
        #       because Pandas allows mixed string column types (and actually has
        #       problems w/ allowing non-string types in object columns)

# Pandas 1.0 has StringDtype extension type
    if col_dtype.name == 'string':
        return ColumnInfo(unicode_dtype)

    if col_dtype == 'bool':
        return ColumnInfo(np.uint8, repr=np.dtype('bool'))

    if col_dtype == np.dtype("O"):
        # Note: this does a full scan of the column... not sure what else to do here
        #       because Pandas allows mixed string column types (and actually has
        #       problems w/ allowing non-string types in object columns)
        inferred_dtype = pd.api.types.infer_dtype(col)

        if inferred_dtype == 'bytes':
            return ColumnInfo(np.bytes_)

        elif inferred_dtype == 'string':
            # TODO we need to make sure this is actually convertible
            return ColumnInfo(unicode_dtype)

        elif inferred_dtype == 'mixed':
            raise ValueError(
                "Column '{}' has mixed value dtype and cannot yet be stored as a TileDB attribute"
            )

    raise ValueError(
        "Unhandled column type: '{}'".format(
            col_dtype
        )