How to use the pmdarima.compat.DTYPE function in pmdarima

if an ``ARIMA`` is fit on exogenous features, it must be provided
            exogenous features for making predictions.

        **fit_kwargs : keyword args
            Extra keyword arguments used for each stage's ``fit`` stage.
            Similar to scikit-learn pipeline keyword args, the keys are
            compound, comprised of the stage name and the argument name
            separated by a "__". For instance, if fitting an ARIMA in stage
            "arima", your kwargs may resemble::

                {"arima__maxiter": 10}
        """
        # Shallow copy
        steps = self.steps_ = self._validate_steps()

        yt = check_endog(y, dtype=DTYPE, copy=False)
        Xt = exogenous
        named_kwargs = self._get_kwargs(**fit_kwargs)

        # store original shape for later in-sample preds
        self.n_samples_ = yt.shape[0]

        for step_idx, name, transformer in self._iter(with_final=False):
            cloned_transformer = clone(transformer)
            kwargs = named_kwargs[name]
            yt, Xt = cloned_transformer.fit_transform(yt, Xt, **kwargs)

            # Replace the transformer of the step with the fitted
            # transformer.
            steps[step_idx] = (name, cloned_transformer)

        # Now fit the final estimator

def load_airpassengers(as_series=False, dtype=DTYPE):
    """Monthly airline passengers.

    The classic Box & Jenkins airline data. Monthly totals of international
    airline passengers, 1949 to 1960.

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If False, will return a 1d
        numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

    Returns

def load_lynx(as_series=False, dtype=DTYPE):
    """Annual numbers of lynx trappings for 1821–1934 in Canada.

    This time-series records the number of skins of predators (lynx) that were
    collected over several years by the Hudson's Bay Company. The dataset was
    taken from Brockwell & Davis (1991) and appears to be the series
    considered by Campbell & Walker (1977).

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If True, the index will be set to
        the observed years. If False, will return a 1d numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

def load_heartrate(as_series=False, dtype=DTYPE):
    """Uniform heart-rate data.

    A sample of heartrate data borrowed from an
    `MIT database `_. The sample consists
    of 150 evenly spaced (0.5 seconds) heartrate measurements.

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If False, will return a 1d
        numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

def load_wineind(as_series=False, dtype=DTYPE):
    """Australian total wine sales by wine makers in bottles <= 1 litre.

    This time-series records wine sales by Australian wine makers between
    Jan 1980 -- Aug 1994. This dataset is found in the R ``forecast`` package.

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If True, the index will be set to
        the observed years/months. If False, will return a 1d numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

    Notes

def load_woolyrnq(as_series=False, dtype=DTYPE):
    """Quarterly production of woollen yarn in Australia.

    This time-series records the quarterly production (in tonnes) of woollen
    yarn in Australia between Mar 1965 and Sep 1994.

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If True, the index will be set to
        the observed years/quarters. If False, will return a 1d numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

    Examples

def _diff_inv_vector(x, lag, differences, xi):
    # R code: if (missing(xi)) xi < - rep(0., lag * differences)
    # R code: if (length(xi) != lag * differences)
    # R code:   stop("'xi' does not have the right length")
    if xi is None:
        xi = np.zeros(lag * differences, dtype=DTYPE)
    else:
        xi = check_endog(xi, dtype=DTYPE, copy=False, force_all_finite=False)
        if xi.shape[0] != lag * differences:
            raise IndexError('"xi" does not have the right length')

    if differences == 1:
        return np.asarray(C_intgrt_vec(x=x, xi=xi, lag=lag))
    else:
        # R code: diffinv.vector(diffinv.vector(x, lag, differences - 1L,
        # R code:               diff(xi, lag=lag, differences=1L)),
        # R code:               lag, 1L, xi[1L:lag])
        return diff_inv(
            x=diff_inv(x=x, lag=lag, differences=differences - 1,
                       xi=diff(x=xi, lag=lag, differences=1)),
            lag=lag,
            differences=1,

def load_austres(as_series=False, dtype=DTYPE):
    """Quarterly residential data.

    Numbers (in thousands) of Australian residents measured quarterly from
    March 1971 to March 1994.

    Parameters
    ----------
    as_series : bool, optional (default=False)
        Whether to return a Pandas series. If False, will return a 1d
        numpy array.

    dtype : type, optional (default=np.float64)
        The type to return for the array. Default is np.float64, which is used
        throughout the package as the default type.

    Returns

def _diff_inv_matrix(x, lag, differences, xi):
    n, m = x.shape
    y = np.zeros((n + lag * differences, m), dtype=DTYPE)

    if m >= 1:  # todo: R checks this. do we need to?
        # R: if(missing(xi)) xi <- matrix(0.0, lag*differences, m)
        if xi is None:
            xi = np.zeros((lag * differences, m), dtype=DTYPE)
        else:
            xi = check_array(
                xi, dtype=DTYPE, copy=False, force_all_finite=False,
                ensure_2d=True)
            if xi.shape != (lag * differences, m):
                raise IndexError('"xi" does not have the right shape')

        # TODO: can we vectorize?
        for i in range(m):
            y[:, i] = _diff_inv_vector(x[:, i], lag, differences, xi[:, i])

    return y

def _diff_inv_vector(x, lag, differences, xi):
    # R code: if (missing(xi)) xi < - rep(0., lag * differences)
    # R code: if (length(xi) != lag * differences)
    # R code:   stop("'xi' does not have the right length")
    if xi is None:
        xi = np.zeros(lag * differences, dtype=DTYPE)
    else:
        xi = check_endog(xi, dtype=DTYPE, copy=False, force_all_finite=False)
        if xi.shape[0] != lag * differences:
            raise IndexError('"xi" does not have the right length')

    if differences == 1:
        return np.asarray(C_intgrt_vec(x=x, xi=xi, lag=lag))
    else:
        # R code: diffinv.vector(diffinv.vector(x, lag, differences - 1L,
        # R code:               diff(xi, lag=lag, differences=1L)),
        # R code:               lag, 1L, xi[1L:lag])
        return diff_inv(
            x=diff_inv(x=x, lag=lag, differences=differences - 1,
                       xi=diff(x=xi, lag=lag, differences=1)),
            lag=lag,
            differences=1,
            xi=xi[:lag]  # R: xi[1L:lag]
        )