# Copyright (C) 2016  Christopher M. Biwer
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 3 of the License, or (at your
# option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General
# Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.


#
# =============================================================================
#
#                                   Preamble
#
# =============================================================================
#
"""
This modules provides functions for calculating the autocorrelation function
and length of a data series.
"""

import numpy
from pycbc.filter.matchedfilter import correlate
from pycbc.types import FrequencySeries, TimeSeries, zeros

def calculate_acf(data, delta_t=1.0, unbiased=False):
    r"""Calculates the one-sided autocorrelation function.

    Calculates the autocorrelation function (ACF) and returns the one-sided
    ACF. The ACF is defined as the autocovariance divided by the variance. The
    ACF can be estimated using

    .. math::

        \hat{R}(k) = \frac{1}{n \sigma^{2}} \sum_{t=1}^{n-k} \left( X_{t} - \mu \right) \left( X_{t+k} - \mu \right)

    Where :math:`\hat{R}(k)` is the ACF, :math:`X_{t}` is the data series at
    time t, :math:`\mu` is the mean of :math:`X_{t}`, and :math:`\sigma^{2}` is
    the variance of :math:`X_{t}`.

    Parameters
    -----------
    data : TimeSeries or numpy.array
        A TimeSeries or numpy.array of data.
    delta_t : float
        The time step of the data series if it is not a TimeSeries instance.
    unbiased : bool
        If True the normalization of the autocovariance function is n-k
        instead of n. This is called the unbiased estimation of the
        autocovariance. Note that this does not mean the ACF is unbiased.

    Returns
    -------
    acf : numpy.array
        If data is a TimeSeries then acf will be a TimeSeries of the
        one-sided ACF. Else acf is a numpy.array.
    """

    # if given a TimeSeries instance then get numpy.array
    if isinstance(data, TimeSeries):
        y = data.numpy()
        delta_t = data.delta_t
    else:
        y = data

    # Zero mean
    y = y - y.mean()
    ny_orig = len(y)

    npad = 1
    while npad < 2*ny_orig:
        npad = npad << 1
    ypad = numpy.zeros(npad)
    ypad[:ny_orig] = y

    # FFT data minus the mean
    fdata = TimeSeries(ypad, delta_t=delta_t).to_frequencyseries()

    # correlate
    # do not need to give the congjugate since correlate function does it
    cdata = FrequencySeries(zeros(len(fdata), dtype=fdata.dtype),
                           delta_f=fdata.delta_f, copy=False)
    correlate(fdata, fdata, cdata)

    # IFFT correlated data to get unnormalized autocovariance time series
    acf = cdata.to_timeseries()
    acf = acf[:ny_orig]

    # normalize the autocovariance
    # note that dividing by acf[0] is the same as ( y.var() * len(acf) )
    if unbiased:
        acf /= ( y.var() * numpy.arange(len(acf), 0, -1) )
    else:
        acf /= acf[0]

    # return input datatype
    if isinstance(data, TimeSeries):
        return TimeSeries(acf, delta_t=delta_t)
    else:
        return acf


def calculate_acl(data, m=5, dtype=int):
    r"""Calculates the autocorrelation length (ACL).

    Given a normalized autocorrelation function :math:`\rho[i]` (by normalized,
    we mean that :math:`\rho[0] = 1`), the ACL :math:`\tau` is:

    .. math::

        \tau = 1 + 2 \sum_{i=1}^{K} \rho[i].

    The number of samples used :math:`K` is found by using the first point
    such that:

    .. math::

        m \tau[K] \leq K,

    where :math:`m` is a tuneable parameter (default = 5). If no such point
    exists, then the given data set it too short to estimate the ACL; in this
    case ``inf`` is returned.

    This algorithm for computing the ACL is taken from:

    N. Madras and A.D. Sokal, J. Stat. Phys. 50, 109 (1988).

    Parameters
    -----------
    data : TimeSeries or array
        A TimeSeries of data.
    m : int
        The number of autocorrelation lengths to use for determining the window
        size :math:`K` (see above).
    dtype : int or float
        The datatype of the output. If the dtype was set to int, then the
        ceiling is returned.

    Returns
    -------
    acl : int or float
        The autocorrelation length. If the ACL cannot be estimated, returns
        ``numpy.inf``.
    """

    # sanity check output data type
    if dtype not in [int, float]:
        raise ValueError("The dtype must be either int or float.")

    # if we have only a single point, just return 1
    if len(data) < 2:
        return 1

    # calculate ACF that is normalized by the zero-lag value
    acf = calculate_acf(data)

    cacf = 2 * acf.numpy().cumsum() - 1
    win = m * cacf <= numpy.arange(len(cacf))
    if win.any():
        acl = cacf[numpy.where(win)[0][0]]
        if dtype == int:
            acl = int(numpy.ceil(acl))
    else:
        acl = numpy.inf
    return acl