flake fixes

        return resid

    def simulate_data(self, num_samples=1000, num_realisations=1, use_cov=True):

        num_sources = self.nsignals

        # Preallocate output

import fractions

from dataclasses import dataclass

from functools import wraps

from copy import deepcopy

import numpy as np

from scipy import fft as sp_fft

def _proc_get_freq_inds(freqvals, fmin, fmax):

    """Get indices of specified frequencies within range."""

    fidx = (freqvals >= fmin) & \

           (freqvals <= fmax)

    return fidx

def _proc_get_time_range(nperseg, nstep, fs, xlen):

    """Get time indices for a given STFT."""

    # Create time window vector

    noverlap = nperseg - nstep

    return np.arange(nperseg/2, xlen - nperseg/2 + 1, nperseg - noverlap)/float(fs)

    """

    if fmin >= fmax:

        logging.error('Selected fmin ({}) is larger than fmax ({})'.format(fmin,))

        logging.error('Selected fmin ({}) is larger than fmax ({})'.format(fmin, fmax))

    logging.info('Setting data within range {0} - {1} to NaN'.format(fmin, fmax))

    fidx = _proc_get_freq_inds(freqvals, fmin, fmax)

    """

    if fmin is None and fmax is None:

        # Just passing through

        return freqs, result

        return freqvals, result

    if fmin >= fmax:

        logging.error('Selected fmin ({}) is larger than fmax ({})'.format(fmin,))

        logging.error('Selected fmin ({}) is larger than fmax ({})'.format(fmin, fmax))

    logging.info('Trimming freq axis to range {0} - {1}'.format(fmin, fmax))

    fmin = freqvals[0] if fmin is None else fmin

    p : array_like

        Array to compute average across

    average : {'mean', 'median', 'median_bias', 'min'}

        Method of 

        Method of averaging

    axis : int

        Index of dimension to average across (default = 0)

    keepdims : bool, optional

        p = np.nanmedian(p, axis=axis, keepdims=keepdims) / bias

    elif average == 'min':

        p = np.nanmin(p, axis=axis, keepdims=keepdims)

    elif average == None:

    elif average is None:

        pass

    else:

        msg = "'average' value of '{0}' not recognised - please use 'mean' or 'median'"

# These functions take input data, run the option handling and execute whatever

# computations are needed

@dataclass

class SpectrumResult:

        DC.add_regressor(name='Constant', rtype='Constant')

    for key in reg_categorical.keys():

        logging.debug('Adding Condition : {0}'.format(key))

        # TODO: Need to review this

        # DC.add_regressor(name=key, rtype='Categorical', datainfo=key, codes=[1])

        DC.add_regressor(name=key, rtype='Parametric', datainfo=key, preproc=None)

    for key in reg_ztrans.keys():

    if axis == -1:

        axis = X.ndim - 1

    if X.ndim != 1 and fit_method in ['pinv', 'lstsq']:

        msg = "Data input should be vector for 'pinv' and 'lstsq' fits - data shape {0} was passed in"

    if axis == -1:

        axis = X.ndim - 1

    if X.ndim != 1 and fit_method in ['pinv', 'lstsq']:

        msg = "Data input should be vector for 'pinv' and 'lstsq' fits - data shape {0} was passed in"

        logging.error(msg.format(X.shape))

        logging.error("Use fit_method='glmtools' for multdimensional data")

        raise ValueError("Fit methods 'pinv' and 'lstsq' not implemented for multidimensional data")

    # Set configuration

    logging.info('Setting config options')

    config = MultiTaperConfig(X.shape[axis],

# 3. Quick

# ISSUES WITH IRASA

# 1. Resampling pairs do nothing & geometric mean within pairs does very little, better that each contributes to a robust median

# 1. Resampling pairs do nothing & geometric mean within pairs does very little,

#    better that each contributes to a robust median

# 2. Fraction inversion resampling factors are more effective on the high end than the low end.

#    They systematically leave a residue on the low frequency end of broad peaks

# 2b.  Though the factors are symmetric in the time domain, oscillatory peaks are irregularly spaced in the frequency domain

# 3. IRASA estimates the aperodic component whilst the oscillatory component is a residual and is therefore mixed with noise.

# 2b.  Though the factors are symmetric in the time domain,

#      oscillatory peaks are irregularly spaced in the frequency domain

# 3. IRASA estimates the aperodic component whilst the oscillatory component is a

#    residual and is therefore mixed with noise.

# 4. Oscillatory, but not aperiodic, component can be negative - neither should ever be negative.

# Solutions

# 1. Resampling is carried out one at a time and all resamples contribute to median where they benefit robust estimation

# 2. Resampling factors are logarithmically spaced - ensuring that oscillatory peaks are (nearly) uniformly spaced in freq-domain

# 3. The final aperiodic component is ap(f) = min(ap(f), full(f)) - this forces aperiodic and oscillatory components to be +ve

# 1. Resampling is carried out one at a time and all resamples contribute

#    to median where they benefit robust estimation

# 2. Resampling factors are logarithmically spaced - ensuring that oscillatory

#     peaks are (nearly) uniformly spaced in freq-domain

# 3. The final aperiodic component is ap(f) = min(ap(f), full(f)) -

#    this forces aperiodic and oscillatory components to be +ve

#    (This constraint should not be enforced for GLM-IRASA)

# ISSUE WITH 1/f fitting

# 1. Unreasonable to assume that one single slope fits whole spectrum

# 1b. This relates to the necessity for a 'knee' in fooof and spectral-plateau as well as lots of empirical results

# 1c. Also need to acknoweldge fundamental ambiguity in solution, particularly where there are broad spans of oscillatory peaks

# 1c. Also need to acknoweldge fundamental ambiguity in solution,

#     particularly where there are broad spans of oscillatory peaks

# Solutions

# 1. 1/f slope is parameterised for purpose such as group/condition comparison or correlation

# 2. Estimate slope in flexible manner (IRASA) that allows for some flexiblity in range of slopes (not perfect though)

# 3. Compute contrasts or correlations with GLM-Spectrum in freq-by-freq approach and visualise whole spectrum effect within aperiodic component.

# 3. Compute contrasts or correlations with GLM-Spectrum in freq-by-freq

#    approach and visualise whole spectrum effect within aperiodic component.

# IDEAS

# 1. IRASA with multitaper would also expect slope to be self-similar across smoothness parameters?

@set_verbose

def irasa(x, method='original', resample_factors=None, aperiodic_average='median',

          #bootstrap_osc=None, bs_average='mean',

          # Periodogram args

          average='median',

          # STFT window args