fit_model.py

"""
Module with functionalities for fitting atmospheric model spectra.
"""

import os
import math
import warnings

from multiprocessing import Pool, cpu_count

import emcee
import numpy as np

try:
    import pymultinest
except:
    warnings.warn('PyMultiNest could not be imported.')

from species.core import constants
from species.data import database
from species.read import read_model, read_object
from species.util import read_util


def lnprior(param,
            bounds,
            modelpar,
            prior=None):
    """
    Internal function for the prior probability.

    Parameters
    ----------
    param : numpy.ndarray
        Parameter values.
    bounds : dict
        Parameter boundaries.
    modelpar : list(str, )
        Parameter names.
    prior : tuple(str, float, float), None
        Gaussian prior on one of the parameters. Currently only possible for the mass, e.g.
        ('mass', 13., 3.) for an expected mass of 13 Mjup with an uncertainty of 3 Mjup. Not
        used if set to None.

    Returns
    -------
    float
        Log prior probability.
    """

    if prior is not None:

        modeldict = {}
        for i, item in enumerate(modelpar):
            modeldict[item] = param[i]

    ln_prior = 0

    for i, item in enumerate(modelpar):

        if bounds[item][0] <= param[i] <= bounds[item][1]:

            if prior is not None and prior[0] == 'mass' and item == 'logg':
                mass = read_util.get_mass(modeldict)
                ln_prior += -0.5*(mass-prior[1])**2/prior[2]**2

            else:
                ln_prior += 0.

        else:
            ln_prior = -np.inf
            break

    return ln_prior


def lnlike(param,
           modelpar,
           objphot,
           distance,
           spectrum,
           modelphot,
           modelspec):
    """
    Internal function for the likelihood function.

    Parameters
    ----------
    param : numpy.ndarray
        Parameter values.
    modelpar : list(str, )
        Parameter names.
    objphot : list(tuple(float, float), )
        List with the fluxes and uncertainties of the object.
    distance : tuple(float, float)
        Distance and uncertainty (pc).
    spectrum : dict
        Dictionary with the spectrum stored as wavelength (um), flux (W m-2 um-1),
        and error (W m-2 um-1), and optionally the covariance matrix and the inverse of
        the covariance matrix.
    modelphot : list(species.read.read_model.ReadModel, )
        List with the interpolated synthetic photometry.
    modelspec : list(species.read.read_model.ReadModel, )
        List with the interpolated synthetic spectra.

    Returns
    -------
    float
        Log likelihood probability.
    """

    paramdict = {}
    spec_scaling = {}

    for i, item in enumerate(modelpar):
        if item == 'radius':
            radius = param[i]

        elif item[:8] == 'scaling_' and item[8:] in spectrum:
            spec_scaling[item[8:]] = param[i]

        else:
            paramdict[item] = param[i]

    scaling = (radius*constants.R_JUP)**2 / (distance[0]*constants.PARSEC)**2

    chisq = 0.

    if objphot is not None:
        for i, item in enumerate(objphot):
            flux = scaling * modelphot[i].spectrum_interp(list(paramdict.values()))
            chisq += (item[0]-flux)**2 / item[1]**2

    if spectrum is not None:
        for i, item in enumerate(spectrum.keys()):
            flux = scaling * modelspec[i].spectrum_interp(list(paramdict.values()))[0, :]

            if item in spec_scaling:
                flux_obs = spec_scaling[item]*spectrum[item][0][:, 1]
            else:
                flux_obs = spectrum[item][0][:, 1]

            if spectrum[item][2] is not None:
                spec_diff = flux_obs - flux
                chisq += np.dot(spec_diff, np.dot(spectrum[item][2], spec_diff))

            else:
                chisq += np.nansum((flux_obs-flux)**2 / spectrum[item][0][:, 2]**2)

    return -0.5*chisq


def lnprob(param,
           bounds,
           modelpar,
           objphot,
           distance,
           prior,
           spectrum,
           modelphot,
           modelspec):
    """
    Internal function for the posterior probability.

    Parameters
    ----------
    param : numpy.ndarray
        Parameter values.
    bounds : dict
        Parameter boundaries.
    modelpar : list(str, )
        Parameter names.
    objphot : list(tuple(float, float), )
        List with the fluxes and uncertainties of the object.
    distance : tuple(float, float)
        Distance and uncertainty (pc).
    prior : tuple(str, float, float)
        Gaussian prior on one of the parameters. Currently only possible for the mass, e.g.
        ('mass', 13., 3.) for an expected mass of 13 Mjup with an uncertainty of 3 Mjup. Not
        used if set to None.
    spectrum : dict
        Wavelength (um), apparent flux (W m-2 um-1), and flux error (W m-2 um-1).
    modelphot : list(species.read.read_model.ReadModel, )
        List with the interpolated synthetic fluxes.
    modelspec : list(species.read.read_model.ReadModel, )
        List with the interpolated synthetic spectra.

    Returns
    -------
    float
        Log posterior probability.
    """

    ln_prior = lnprior(param, bounds, modelpar, prior)

    if math.isinf(ln_prior):
        ln_prob = -np.inf

    else:
        ln_prob = ln_prior + lnlike(param,
                                    modelpar,
                                    objphot,
                                    distance,
                                    spectrum,
                                    modelphot,
                                    modelspec)

    if np.isnan(ln_prob):
        ln_prob = -np.inf

    return ln_prob


class FitModel:
    """
    Class for fitting atmospheric model spectra to photometric and/or spectroscopic data.
    """

    def __init__(self,
                 object_name,
                 filters,
                 model,
                 bounds=None,
                 inc_phot=True,
                 inc_spec=True):
        """
        For each photometric point and spectrum, the model grid is linearly interpolated at the
        required synthetic photometry and wavelength sampling before running the MCMC. Therefore,
        the computation time of this initial interpolation depends on the wavelength range and
        spectral resolution of the spectra that are stored in the database, and the prior
        boundaries that are chosen with ``bounds``.

        Parameters
        ----------
        object_name : str
            Object name in the database as created with
            :func:`~species.data.database.Database.add_object` or
            :func:`~species.data.database.Database.add_companion`.
        filters : tuple(str, )
            Filter names for which the photometry is selected. All available photometry of the
            object is selected if set to None.
        model : str
            Atmospheric model (e.g. 'drift-phoenix', 'petitcode-cool-cloudy', or 'bt-settl').
        bounds : dict, None
            Parameter boundaries. The full range is used for each parameter if set to None. In that
            case, the radius range is set to 0.8-1.5 Rjup. It is also possible to specify the bounds
            for a subset of the parameters, for example, ``{'radius': (0.5, 10.)}``. Restricting
            the boundaries will decrease the computation time with the interpolation prior to the
            MCMC sampling. An additional scaling parameter can be fitted for each spectrum in which
            case, the boundaries have to be provided with the database tag of the spectrum.
            For example, ``{'sphere_ifs': (0.5, 2.)}`` if the spectrum is stored as ``sphere_ifs``
            with :func:`~species.data.database.Database.add_object`.
        inc_phot : bool
            Include photometric data in the fit.
        inc_spec : bool
            Include spectroscopic data in the fit.

        Returns
        -------
        NoneType
            None
        """

        self.object = read_object.ReadObject(object_name)
        self.distance = self.object.get_distance()

        self.model = model
        self.bounds = bounds

        if not inc_phot and not inc_spec:
            raise ValueError('No photometric or spectral data has been selected.')

        if self.bounds is not None:
            readmodel = read_model.ReadModel(self.model)
            bounds_grid = readmodel.get_bounds()

            for item in bounds_grid:
                if item not in self.bounds:
                    self.bounds[item] = bounds_grid[item]

        else:
            readmodel = read_model.ReadModel(self.model, None, None)
            self.bounds = readmodel.get_bounds()

        if 'radius' not in self.bounds:
            self.bounds['radius'] = (0.8, 1.5)

        print('Prior and interpolation boundaries:')
        for key, value in self.bounds.items():
            print(f'   - {key} = {value}')

        if inc_phot:
            self.objphot = []
            self.modelphot = []

            if filters is None:
                species_db = database.Database()
                objectbox = species_db.get_object(object_name, None)
                filters = objectbox.filters

            for item in filters:
                print(f'Interpolating {item}...', end='', flush=True)

                readmodel = read_model.ReadModel(self.model, filter_name=item)
                readmodel.interpolate_grid(self.bounds)
                self.modelphot.append(readmodel)

                print(f' [DONE]')

                obj_phot = self.object.get_photometry(item)
                self.objphot.append((obj_phot[2], obj_phot[3]))

        else:
            self.objphot = None
            self.modelphot = None

        if inc_spec:
            self.spectrum = self.object.get_spectrum()

            self.modelspec = []
            for key, value in self.spectrum.items():
                print(f'\rInterpolating {key}...', end='', flush=True)

                wavel_range = (0.9*value[0][0, 0], 1.1*value[0][-1, 0])

                readmodel = read_model.ReadModel(self.model, wavel_range=wavel_range)

                readmodel.interpolate_grid(self.bounds,
                                           wavel_resample=self.spectrum[key][0][:, 0],
                                           smooth=True,
                                           spec_res=self.spectrum[key][3])

                self.modelspec.append(readmodel)

                print(f' [DONE]')

        else:
            self.spectrum = None
            self.modelspec = None

        self.modelpar = readmodel.get_parameters()
        self.modelpar.append('radius')

        if self.spectrum is not None:
            for item in self.spectrum:
                if item in bounds:
                    self.modelpar.append(f'scaling_{item}')
                    self.bounds[f'scaling_{item}'] = (bounds[item][0], bounds[item][1])
                    del self.bounds[item]

    def run_mcmc(self,
                 tag,
                 guess,
                 nwalkers=200,
                 nsteps=1000,
                 prior=None):
        """
        Function to run the MCMC sampler of ``emcee``.

        Parameters
        ----------
        tag : str
            Database tag where the samples will be stored.
        guess : dict, None
            Guess for the parameter values. Random values between the boundary values are used
            if set to None.
        nwalkers : int
            Number of walkers.
        nsteps : int
            Number of steps per walker.
        prior : tuple(str, float, float), None
            Gaussian prior on one of the parameters. Currently only possible for the mass, e.g.
            ('mass', 13., 3.) for an expected mass of 13 Mjup with an uncertainty of 3 Mjup. Not
            used if set to None.

        Returns
        -------
        NoneType
            None
        """

        sigma = {'teff': 5., 'logg': 0.01, 'feh': 0.01, 'fsed': 0.01, 'co': 0.01, 'radius': 0.01}

        if self.spectrum is not None:
            for item in self.spectrum:
                if f'scaling_{item}' in self.bounds:
                    sigma[f'scaling_{item}'] = 0.01
                    guess[f'scaling_{item}'] = guess[item]
                    del guess[item]

        print('Running MCMC...')

        ndim = len(self.bounds)

        initial = np.zeros((nwalkers, ndim))
        for i, item in enumerate(self.modelpar):
            if guess[item] is not None:
                initial[:, i] = guess[item] + np.random.normal(0, sigma[item], nwalkers)

            else:
                initial[:, i] = np.random.uniform(low=self.bounds[item][0],
                                                  high=self.bounds[item][1],
                                                  size=nwalkers)

        with Pool(processes=cpu_count()):
            ens_sampler = emcee.EnsembleSampler(nwalkers,
                                                ndim,
                                                lnprob,
                                                args=([self.bounds,
                                                       self.modelpar,
                                                       self.objphot,
                                                       self.distance,
                                                       prior,
                                                       self.spectrum,
                                                       self.modelphot,
                                                       self.modelspec]))

            ens_sampler.run_mcmc(initial, nsteps, progress=True)

        species_db = database.Database()

        spec_labels = []

        if self.spectrum is not None:
            for item in self.spectrum:
                if f'scaling_{item}' in self.bounds:
                    spec_labels.append(f'scaling_{item}')

        else:
            spec_labels = None

        species_db.add_samples(sampler='emcee',
                               samples=ens_sampler.chain,
                               ln_prob=ens_sampler.lnprobability,
                               mean_accept=np.mean(ens_sampler.acceptance_fraction),
                               spectrum=('model', self.model),
                               tag=tag,
                               modelpar=self.modelpar,
                               distance=self.distance[0],
                               spec_labels=spec_labels)

    def run_multinest(self,
                      tag,
                      n_live_points=4000,
                      output='multinest/'):
        """
        Function to run the ``PyMultiNest`` wrapper of the ``MultiNest`` sampler. While
        ``PyMultiNest`` can be installed with ``pip`` from the PyPI repository, ``MultiNest``
        has to to be build manually. See the ``PyMultiNest`` documentation for details:
        http://johannesbuchner.github.io/PyMultiNest/install.html. Note that the library path
        of ``MultiNest`` should be set to the environmental variable ``LD_LIBRARY_PATH`` on a
        Linux machine and ``DYLD_LIBRARY_PATH`` on a Mac. Alternatively, the variable can be
        set before importing the ``species`` package, for example:

        .. code-block:: python

            >>> import os
            >>> os.environ['DYLD_LIBRARY_PATH'] = '/path/to/MultiNest/lib'
            >>> import species

        Parameters
        ----------
        tag : str
            Database tag where the samples will be stored.
        n_live_points : int
            Number of live points.
        output : str
            Path that is used for the output files from MultiNest.

        Returns
        -------
        NoneType
            None
        """

        print('Running nested sampling...')

        # create the output folder if required

        if not os.path.exists(output):
            os.mkdir(output)

        # create a dictionary with the cube indices of the parameters

        cube_index = {}
        for i, item in enumerate(self.modelpar):
            cube_index[item] = i

        def lnprior_multinest(cube, n_dim, n_param):
            """
            Function to transform the unit cube into the parameter cube. It is not clear how to
            pass additional arguments to the function, therefore it is placed here.

            Parameters
            ----------
            cube : pymultinest.run.LP_c_double
                Unit cube.

            Returns
            -------
            NoneType
                None
            """

            # Effective temperature (K)
            cube[cube_index['teff']] = self.bounds['teff'][0] + \
                (self.bounds['teff'][1]-self.bounds['teff'][0])*cube[cube_index['teff']]

            # Surface gravity (dex)
            cube[cube_index['logg']] = self.bounds['logg'][0] + \
                (self.bounds['logg'][1]-self.bounds['logg'][0])*cube[cube_index['logg']]

            # Radius (Rjup)
            cube[cube_index['radius']] = self.bounds['radius'][0] + \
                (self.bounds['radius'][1]-self.bounds['radius'][0])*cube[cube_index['radius']]

            # Metallicity [Fe/H]
            if 'feh' in self.bounds:
                cube[cube_index['feh']] = self.bounds['feh'][0] + \
                    (self.bounds['feh'][1]-self.bounds['feh'][0])*cube[cube_index['feh']]

            # C/O ratio
            if 'co' in self.bounds:
                cube[cube_index['co']] = self.bounds['co'][0] + \
                    (self.bounds['co'][1]-self.bounds['co'][0])*cube[cube_index['co']]

            # Sedimentation parameter
            if 'fsed' in self.bounds:

                cube[cube_index['fsed']] = self.bounds['fsed'][0] + \
                    (self.bounds['fsed'][1]-self.bounds['fsed'][0])*cube[cube_index['fsed']]

            # Spectrum scaling
            if self.spectrum is not None:
                for item in self.spectrum:
                    if f'scaling_{item}' in self.bounds:
                        cube[cube_index[f'scaling_{item}']] = self.bounds[f'scaling_{item}'][0] + \
                            (self.bounds[f'scaling_{item}'][1]-self.bounds[f'scaling_{item}'][0]) \
                            * cube[cube_index[f'scaling_{item}']]

        def lnlike_multinest(cube, n_dim, n_param):
            """
            Function for the logarithm of the likelihood, computed from the parameter cube.

            Parameters
            ----------
            cube : pymultinest.run.LP_c_double
                Unit cube.

            Returns
            -------
            float
                The logarithm of the likelihood.
            """

            paramdict = {}
            spec_scaling = {}

            for i, item in enumerate(self.modelpar):
                if item == 'radius':
                    radius = cube[cube_index['radius']]

                elif item[:8] == 'scaling_' and item[8:] in self.spectrum:
                    spec_scaling[item[8:]] = cube[cube_index[item]]

                else:
                    paramdict[item] = cube[cube_index[item]]

            scaling = (radius*constants.R_JUP)**2 / (self.distance[0]*constants.PARSEC)**2

            chisq = 0.

            if self.objphot is not None:
                for i, item in enumerate(self.objphot):
                    flux = scaling * self.modelphot[i].spectrum_interp(list(paramdict.values()))
                    chisq += (item[0]-flux)**2 / item[1]**2

            if self.spectrum is not None:
                for i, item in enumerate(self.spectrum.keys()):

                    flux = scaling * self.modelspec[i].spectrum_interp(list(paramdict.values()))[0, :]

                    if item in spec_scaling:
                        flux_obs = spec_scaling[item]*self.spectrum[item][0][:, 1]
                    else:
                        flux_obs = self.spectrum[item][0][:, 1]

                    if self.spectrum[item][2] is not None:
                        spec_diff = flux_obs - flux
                        chisq += np.dot(spec_diff, np.dot(self.spectrum[item][2], spec_diff))

                    else:
                        chisq += np.nansum((flux_obs-flux)**2 / self.spectrum[item][0][:, 2]**2)

            return -0.5*chisq

        pymultinest.run(lnlike_multinest,
                        lnprior_multinest,
                        len(self.modelpar),
                        outputfiles_basename=output,
                        resume=False,
                        n_live_points=n_live_points)

        samples = np.loadtxt(f'{output}/post_equal_weights.dat')

        species_db = database.Database()

        spec_labels = []

        if self.spectrum is not None:
            for item in self.spectrum:
                if f'scaling_{item}' in self.bounds:
                    spec_labels.append(f'scaling_{item}')

        else:
            spec_labels = None

        species_db.add_samples(sampler='multinest',
                               samples=samples[:, :-1],
                               ln_prob=samples[:, -1],
                               mean_accept=None,
                               spectrum=('model', self.model),
                               tag=tag,
                               modelpar=self.modelpar,
                               distance=self.distance[0],
                               spec_labels=spec_labels)