Performance of classifiers

Classifier for HasNS trained on \(m_2 \leq 3 M_{\odot}\) to have an NS. 2H EoS used for remnant matter computation.

import pickle

import pkg_resources

import numpy as np
import matplotlib.pyplot as plt

from astropy.table import Table

from ligo import em_bright

with pkg_resources.resource_stream('ligo.data', 'knn_ns_classifier.pkl') as f:
    clf_ns = pickle.load(f)

with pkg_resources.resource_stream('ligo.data', 'knn_em_classifier.pkl') as f:
    clf_em = pickle.load(f)

# sweep masses
mass1 = np.linspace(1, 100, 1000)
mass2 = np.linspace(1, 100, 1000)

t = Table(data=np.vstack((np.repeat(mass1, mass2.size),
                          np.tile(mass2, mass1.size))).T, names=('mass1', 'mass2'))

mask = t['mass1'] > t['mass2']

t = t[mask]

# vary spin, fix SNR
spins = Table(data=np.vstack((np.repeat(np.linspace(0, 1, 2), 2),
              np.tile(np.linspace(0, 1, 2), 2))).T,
              names=('chi1', 'chi2'))

def make_plots(features, predictions, title, fig_idx):
    fig_, idx = fig_idx
    fig_.add_subplot(4, 1, idx)
    # indices 0 and 1 correspond to mass1 and mass2 respectively
    plt.scatter(features.T[0], features.T[1],
                s=10, c=predictions)
    plt.title(title)
    plt.tight_layout()
    plt.colorbar(label='Probability')
    # plot chirp mass contours
    m1 = np.linspace(1, 50, 100)
    m2 = np.linspace(1, 50, 100)
    M1, M2 = np.meshgrid(m1, m2)
    Mc = (M1*M2)**(3./5.)/(M1 + M2)**(1./5.)
    Mc = np.tril(Mc).T

    CS = plt.contour(M1, M2, Mc, levels=[5, 6, 7, 8, 9],
                     colors='black', linewidths=1)

    plt.clabel(CS, inline=True, fontsize=16)
    plt.xlim((1, 50))
    plt.ylim((1, 14))
    plt.axhline(y=3.0, c='r')
    plt.xlabel(r'$m_1$', fontsize=16)
    plt.ylabel(r'$m_2$', fontsize=16)
    plt.title(title)
    plt.tight_layout()

Parameter sweep for the HasNS predictor

fig = plt.figure(figsize=(14, 16))

for idx, spin_vals in enumerate(spins):
    SNR = 10.
    title = "chi1 = {0}; chi2 = {1}; SNR = {2}".format(spin_vals['chi1'],
                                                       spin_vals['chi2'],
                                                       SNR)
    SNR *= np.ones(np.sum(mask))
    chi1 = spin_vals['chi1'] * np.ones(np.sum(mask))
    chi2 = spin_vals['chi2'] * np.ones(np.sum(mask))

    # make predictions and make plots
    param_sweep_features = np.stack([t['mass1'], t['mass2'], chi1, chi2, SNR]).T
    ns_predictions = clf_ns.predict_proba(param_sweep_features).T[1]
    make_plots(param_sweep_features, ns_predictions, title, (fig, idx+1))

Parameter sweep for the HasRemnant predictor

fig = plt.figure(figsize=(14, 16))

for idx, spin_vals in enumerate(spins):
    SNR = 10.
    title = "chi1 = {0}; chi2 = {1}; SNR = {2}".format(spin_vals['chi1'],
                                                       spin_vals['chi2'],
                                                       SNR)
    SNR *= np.ones(np.sum(mask))
    chi1 = spin_vals['chi1'] * np.ones(np.sum(mask))
    chi2 = spin_vals['chi2'] * np.ones(np.sum(mask))

    # make predictions and make plots
    param_sweep_features = np.stack([t['mass1'], t['mass2'], chi1, chi2, SNR]).T
    remnant_matter_predictions = clf_em.predict_proba(param_sweep_features).T[1]
    make_plots(param_sweep_features,
               remnant_matter_predictions,
               title,
               (fig, idx+1))

Variation with SNR for HasNS predictor

fig = plt.figure(figsize=(14, 16))

signal_to_noise = [7.0, 8.0, 9.0, 10.0]

for idx, SNR in enumerate(signal_to_noise):
    spin_1 = 0.
    spin_2 = 0.
    title = "chi1 = {0}; chi2 = {1}; SNR = {2}".format(spin_1,
                                                       spin_2,
                                                       SNR)
    SNR *= np.ones(np.sum(mask))
    chi1 = spin_1 * np.ones(np.sum(mask))
    chi2 = spin_2 * np.ones(np.sum(mask))

    # make predictions and make plots
    param_sweep_features = np.stack([t['mass1'], t['mass2'], chi1, chi2, SNR]).T
    ns_predictions = clf_ns.predict_proba(param_sweep_features).T[1]
    make_plots(param_sweep_features, ns_predictions,
               title, (fig, idx+1))

Variation with SNR for HasRemnant predictor

fig = plt.figure(figsize=(14, 16))

signal_to_noise = [7.0, 8.0, 9.0, 10.0]

for idx, SNR in enumerate(signal_to_noise):
    spin_1 = 0.
    spin_2 = 0.
    title = "chi1 = {0}; chi2 = {1}; SNR = {2}".format(spin_1,
                                                       spin_2,
                                                       SNR)
    SNR *= np.ones(np.sum(mask))
    chi1 = spin_1 * np.ones(np.sum(mask))
    chi2 = spin_2 * np.ones(np.sum(mask))

    # make predictions and make plots
    param_sweep_features = np.stack([t['mass1'], t['mass2'], chi1, chi2, SNR]).T
    remnant_matter_predictions = clf_em.predict_proba(param_sweep_features).T[1]
    make_plots(param_sweep_features,remnant_matter_predictions,
               title, (fig, idx+1))