gmm_tools.py

"""@namespace IMP.isd.gmm_tools
   Tools for handling Gaussian Mixture Models.
"""

import IMP
import IMP.core
import IMP.algebra
import IMP.atom
import IMP.em
import numpy as np
import numpy.linalg
import itertools

try:
    import sklearn.mixture  # noqa: F401
    nosklearn = False
except ImportError:
    nosklearn = True
from math import exp, sqrt


def decorate_gmm_from_text(in_fn, ps, mdl, transform=None, radius_scale=1.0,
                           mass_scale=1.0):
    """read the output from write_gmm_to_text, decorate as Gaussian and Mass"""
    ncomp = 0
    added_ps = itertools.count(1)
    with open(in_fn, 'r') as inf:
        for line in inf:
            if line[0] != '#':
                if ncomp > len(ps) - 1:
                    ps.append(IMP.Particle(mdl, "GMM%d" % next(added_ps)))
                p = ps[ncomp]
                fields = line.split('|')
                weight = float(fields[2])
                center = list(map(float, fields[3].split()))
                covar = np.array(
                    list(map(float, fields[4].split()))).reshape((3, 3))
                shape = IMP.algebra.get_gaussian_from_covariance(
                    covar.tolist(), IMP.algebra.Vector3D(center))
                if not IMP.core.Gaussian.get_is_setup(p):
                    g = IMP.core.Gaussian.setup_particle(ps[ncomp], shape)
                else:
                    g = IMP.core.Gaussian(ps[ncomp])
                    g.set_gaussian(shape)
                if not IMP.atom.Mass.get_is_setup(p):
                    IMP.atom.Mass.setup_particle(p, weight * mass_scale)
                else:
                    IMP.atom.Mass(p).set_mass(weight * mass_scale)
                rmax = sqrt(max(g.get_variances())) * radius_scale
                if not IMP.core.XYZR.get_is_setup(ps[ncomp]):
                    IMP.core.XYZR.setup_particle(ps[ncomp], rmax)
                else:
                    IMP.core.XYZR(ps[ncomp]).set_radius(rmax)
                if transform is not None:
                    IMP.core.transform(IMP.core.RigidBody(ps[ncomp]),
                                       transform)
                ncomp += 1


def write_gmm_to_text(ps, out_fn, comments=[]):
    """write a list of gaussians to text. must be decorated as Gaussian
       and Mass"""
    print('will write GMM text to', out_fn)
    with open(out_fn, 'w') as outf:
        for comment in comments:
            outf.write('# ' + comment + '\n')
        outf.write('#|num|weight|mean|covariance matrix|\n')
        for ng, g in enumerate(ps):
            shape = IMP.core.Gaussian(g).get_gaussian()
            weight = IMP.atom.Mass(g).get_mass()
            covar = [c for row in IMP.algebra.get_covariance(shape)
                     for c in row]
            mean = list(shape.get_center())
            fm = [ng, weight] + mean + covar
            outf.write('|{}|{}|{} {} {}|'
                       '{} {} {} {} {} {} {} {} {}|\n'.format(*fm))


def gmm2map(to_draw, voxel_size, bounding_box=None, origin=None, fast=False,
            factor=2.5):
    if type(to_draw[0]) in (IMP.Particle, IMP.atom.Hierarchy,
                            IMP.core.Hierarchy):
        ps = to_draw
    elif type(to_draw[0]) is IMP.core.Gaussian:
        ps = [g.get_particle() for g in to_draw]
    else:
        print('ps must be Particles or Gaussians')
        return
    if bounding_box is None:
        if len(ps) > 1:
            s = IMP.algebra.get_enclosing_sphere(
                [IMP.core.XYZ(p).get_coordinates() for p in ps])
            s2 = IMP.algebra.Sphere3D(s.get_center(), s.get_radius() * 3)
        else:
            g = IMP.core.Gaussian(ps[0]).get_gaussian()
            s2 = IMP.algebra.Sphere3D(g.get_center(),
                                      max(g.get_variances()) * 3)
        bounding_box = IMP.algebra.get_bounding_box(s2)
    shapes = []
    weights = []
    for p in ps:
        shapes.append(IMP.core.Gaussian(p).get_gaussian())
        weights.append(IMP.atom.Mass(p).get_mass())
    print('rasterizing')
    if fast:
        grid = IMP.algebra.get_rasterized_fast(shapes, weights, voxel_size,
                                               bounding_box, factor)
    else:
        grid = IMP.algebra.get_rasterized(shapes, weights, voxel_size,
                                          bounding_box)
    print('creating map')
    d1 = IMP.em.create_density_map(grid)
    if origin is not None:
        d1.set_origin(origin)
    return d1


def write_gmm_to_map(to_draw, out_fn, voxel_size, bounding_box=None,
                     origin=None, fast=False, factor=2.5):
    """write density map from GMM. input can be either particles
       or gaussians"""
    d1 = gmm2map(to_draw, voxel_size, bounding_box, origin, fast)
    print('will write GMM map to', out_fn)
    IMP.em.write_map(d1, out_fn, IMP.em.MRCReaderWriter())
    del d1


def write_sklearn_gmm_to_map(gmm, out_fn, apix=0, bbox=None, dmap_model=None):
    """write density map directly from sklearn GMM (kinda slow) """
    # create density
    if dmap_model is not None:
        d1 = IMP.em.create_density_map(dmap_model)
    else:
        d1 = IMP.em.create_density_map(bbox, apix)

    # fill it with values from the GMM
    print('getting coords')
    nvox = d1.get_number_of_voxels()
    apos = [list(d1.get_location_by_voxel(nv)) for nv in range(nvox)]

    print('scoring')
    scores = gmm.score(apos)

    print('assigning')
    for nv, score in enumerate(scores):
        d1.set_value(nv, exp(score))
    print('will write GMM map to', out_fn)
    IMP.em.write_map(d1, out_fn, IMP.em.MRCReaderWriter())


def draw_points(pts, out_fn,
                trans=IMP.algebra.get_identity_transformation_3d(),
                use_colors=False):
    """ given some points (and optional transform), write them to chimera
        'bild' format; colors flag only applies to ellipses, otherwise it'll
        be weird"""
    with open(out_fn, 'w') as outf:
        # print 'will draw',len(pts),'points'
        # write first point in red
        pt = trans.get_transformed(IMP.algebra.Vector3D(pts[0]))
        start = 0
        if use_colors:
            outf.write('.color 1 0 0\n.dotat %.2f %.2f %.2f\n'
                       % (pt[0], pt[1], pt[2]))
            start = 1

        # write remaining points in green
        if use_colors:
            outf.write('.color 0 1 0\n')
            colors = ['0 1 0', '0 0 1', '0 1 1']
        for nt, t in enumerate(pts[start:]):
            if use_colors and nt % 2 == 0:
                outf.write('.color %s\n' % colors[nt / 2])
            pt = trans.get_transformed(IMP.algebra.Vector3D(t))
            outf.write('.dotat %.2f %.2f %.2f\n' % (pt[0], pt[1], pt[2]))


def create_gmm_for_bead(mdl,
                        particle,
                        n_components,
                        sampled_points=100000,
                        num_iter=100):
    print('fitting bead with', n_components, 'gaussians')
    dmap = IMP.em.SampledDensityMap([particle], 1.0, 1.0,
                                    IMP.atom.Mass.get_mass_key(), 3,
                                    IMP.em.SPHERE)
    IMP.em.write_map(dmap, 'test_intermed.mrc')
    pts = IMP.isd.sample_points_from_density(dmap, sampled_points)
    draw_points(pts, 'pts.bild')
    density_particles = []
    fit_gmm_to_points(pts, n_components, mdl, density_particles, num_iter,
                      'full',
                      mass_multiplier=IMP.atom.Mass(particle).get_mass())
    return density_particles, dmap


def sample_and_fit_to_particles(model,
                                fragment_particles,
                                num_components,
                                sampled_points=1000000,
                                simulation_res=0.5,
                                voxel_size=1.0,
                                num_iter=100,
                                covariance_type='full',
                                multiply_by_total_mass=True,
                                output_map=None,
                                output_txt=None):
    density_particles = []
    if multiply_by_total_mass:
        mass_multiplier = sum((IMP.atom.Mass(p).get_mass()
                               for p in set(fragment_particles)))
        print('add_component_density: will multiply by mass', mass_multiplier)

    # simulate density from ps, then calculate points to fit
    print('add_component_density: sampling points')
    dmap = IMP.em.SampledDensityMap(
        fragment_particles, simulation_res, voxel_size,
        IMP.atom.Mass.get_mass_key(), 3)
    dmap.calcRMS()
    pts = IMP.isd.sample_points_from_density(dmap, sampled_points)

    # fit GMM
    print('add_component_density: fitting GMM to', len(pts), 'points')
    fit_gmm_to_points(points=pts,
                      n_components=num_components,
                      mdl=model,
                      ps=density_particles,
                      num_iter=num_iter,
                      covariance_type=covariance_type,
                      mass_multiplier=mass_multiplier)

    if output_txt is not None:
        write_gmm_to_text(density_particles, output_txt)
    if output_map is not None:
        write_gmm_to_map(to_draw=density_particles,
                         out_fn=output_map,
                         voxel_size=voxel_size,
                         bounding_box=IMP.em.get_bounding_box(dmap))

    return density_particles


def fit_gmm_to_points(points,
                      n_components,
                      mdl,
                      ps=[],
                      num_iter=100,
                      covariance_type='full',
                      min_covar=0.001,
                      init_centers=[],
                      force_radii=-1.0,
                      force_weight=-1.0,
                      mass_multiplier=1.0):
    """fit a GMM to some points. Will return the score and the Akaike score.
    Akaike information criterion for the current model fit. It is a measure
    of the relative quality of the GMM that takes into account the
    parsimony and the goodness of the fit.
    if no particles are provided, they will be created

    points:            list of coordinates (python)
    n_components:      number of gaussians to create
    mdl:               IMP Model
    ps:                list of particles to be decorated. if empty, will add
    num_iter:          number of EM iterations
    covariance_type:   covar type for the gaussians. options: 'full',
                       'diagonal', 'spherical'
    min_covar:         assign a minimum value to covariance term. That is used
                       to have more spherical shaped gaussians
    init_centers:      initial coordinates of the GMM
    force_radii:       fix the radii (spheres only)
    force_weight:      fix the weights
    mass_multiplier:   multiply the weights of all the gaussians by this value
    dirichlet:         use the DGMM fitting (can reduce number of components,
                       takes longer)
    """

    new_sklearn = False
    try:
        from sklearn.mixture import GMM
    except ImportError:
        from sklearn.mixture import GaussianMixture
        new_sklearn = True

    print('creating GMM with n_components', n_components, 'n_iter', num_iter,
          'covar type', covariance_type)
    if new_sklearn:
        # aic() calls size() on points, so it needs to be
        # a numpy array, not a list
        points = np.array(points)
        weights_init = precisions_init = None
        if force_radii != -1.0:
            print('warning: radii can no longer be forced, but setting '
                  'initial values to ', force_radii)
            precisions_init = np.array([[1. / force_radii] * 3
                                       for i in range(n_components)])
        if force_weight != -1.0:
            print('warning: weights can no longer be forced, but setting '
                  'initial values to ', force_weight)
            weights_init = np.array([force_weight] * n_components)

        gmm = GaussianMixture(n_components=n_components,
                              max_iter=num_iter,
                              covariance_type=covariance_type,
                              weights_init=weights_init,
                              precisions_init=precisions_init,
                              means_init=None if init_centers == []
                              else init_centers)
    else:
        params = 'm'
        init_params = 'm'
        if force_radii == -1.0:
            params += 'c'
            init_params += 'c'
        else:
            covariance_type = 'spherical'
            print('forcing spherical with radii', force_radii)

        if force_weight == -1.0:
            params += 'w'
            init_params += 'w'
        else:
            print('forcing weights to be', force_weight)

        gmm = GMM(n_components=n_components, n_iter=num_iter,
                  covariance_type=covariance_type, min_covar=min_covar,
                  params=params, init_params=init_params)
        if force_weight != -1.0:
            gmm.weights_ = np.array([force_weight] * n_components)
        if force_radii != -1.0:
            gmm.covars_ = np.array([[force_radii] * 3
                                    for i in range(n_components)])
        if init_centers != []:
            gmm.means_ = init_centers
    print('fitting')
    model = gmm.fit(points)
    score = gmm.score(points)
    akaikescore = model.aic(points)

    # convert format to core::Gaussian
    if new_sklearn:
        covars = gmm.covariances_
    else:
        covars = gmm.covars_
    for ng in range(n_components):
        covar = covars[ng]
        if covar.size == 3:
            covar = np.diag(covar).tolist()
        else:
            covar = covar.tolist()
        center = list(gmm.means_[ng])
        weight = mass_multiplier * gmm.weights_[ng]
        if ng >= len(ps):
            ps.append(IMP.Particle(mdl))
        shape = IMP.algebra.get_gaussian_from_covariance(
            covar, IMP.algebra.Vector3D(center))
        g = IMP.core.Gaussian.setup_particle(ps[ng], shape)
        IMP.atom.Mass.setup_particle(ps[ng], weight)
        IMP.core.XYZR.setup_particle(ps[ng], sqrt(max(g.get_variances())))

    return (score, akaikescore)


def fit_dirichlet_gmm_to_points(
        points, n_components, mdl, ps=[], num_iter=100,
        covariance_type='full', mass_multiplier=1.0):
    """fit a GMM to some points. Will return core::Gaussians.
    if no particles are provided, they will be created

    points:            list of coordinates (python)
    n_components:      number of gaussians to create
    mdl:               IMP Model
    ps:                list of particles to be decorated. if empty, will add
    num_iter:          number of EM iterations
    covariance_type:   covar type for the gaussians. options: 'full',
                       'diagonal', 'spherical'
    init_centers:      initial coordinates of the GMM
    force_radii:       fix the radii (spheres only)
    force_weight:      fix the weights
    mass_multiplier:   multiply the weights of all the gaussians by this value
    """

    new_sklearn = True
    try:
        from sklearn.mixture import BayesianGaussianMixture
    except ImportError:
        from sklearn.mixture import DPGMM
        new_sklearn = False

    # create and fit GMM
    print('using dirichlet prior')
    if new_sklearn:
        gmm = BayesianGaussianMixture(
            weight_concentration_prior_type='dirichlet_process',
            n_components=n_components, max_iter=num_iter,
            covariance_type=covariance_type)
    else:
        gmm = DPGMM(n_components=n_components, n_iter=num_iter,
                    covariance_type=covariance_type)

    gmm.fit(points)

    # convert format to core::Gaussian
    if not new_sklearn:
        gmm.precisions_ = gmm.precs_
    for ng in range(n_components):
        invcovar = gmm.precisions_[ng]
        covar = np.linalg.inv(invcovar)
        if covar.size == 3:
            covar = np.diag(covar).tolist()
        else:
            covar = covar.tolist()
        center = list(gmm.means_[ng])
        weight = mass_multiplier * gmm.weights_[ng]
        if ng >= len(ps):
            ps.append(IMP.Particle(mdl))
        shape = IMP.algebra.get_gaussian_from_covariance(
            covar, IMP.algebra.Vector3D(center))
        g = IMP.core.Gaussian.setup_particle(ps[ng], shape)
        IMP.atom.Mass.setup_particle(ps[ng], weight)
        IMP.core.XYZR.setup_particle(ps[ng], sqrt(max(g.get_variances())))