def test_known_regression(self):
# prevent regression of known bug
self.assertEqual(mag_to_absmag(1., MAGIC_NUMBER), MAGIC_NUMBER)
self.assertEqual(mag_to_absmag(MAGIC_NUMBER, MAGIC_NUMBER), MAGIC_NUMBER)
self.assertEqual(
np.all(mag_to_absmag(MAGIC_NUMBER, MAGIC_NUMBER, 1.) == (MAGIC_NUMBER, MAGIC_NUMBER)), True)
self.assertEqual(
np.all(mag_to_fakemag(MAGIC_NUMBER, MAGIC_NUMBER, 1.) == (MAGIC_NUMBER, MAGIC_NUMBER)), True)
self.assertEqual(mag_to_fakemag(1., MAGIC_NUMBER), MAGIC_NUMBER)
self.assertEqual(mag_to_fakemag(MAGIC_NUMBER, MAGIC_NUMBER), MAGIC_NUMBER)
self.assertEqual(fakemag_to_pc(1., MAGIC_NUMBER).value, MAGIC_NUMBER)
self.assertEqual(fakemag_to_pc(-1., 2.).value, MAGIC_NUMBER)
self.assertEqual(absmag_to_pc(1., MAGIC_NUMBER).value, MAGIC_NUMBER)
except KeyError:
parallax = apogeegaia_file["parallax"]
parallax_error = apogeegaia_file["parallax_error"]
# set negative parallax after constant offset correction to np.nan
parallax_error[(parallax < 0) & (parallax > 1e10)] = np.nan
parallax[(parallax < 0) & (parallax > 1e10)] = np.nan
# inference
net = load_folder(astronn_dist_model)
# pred, pred_err = net.predict(all_spec)
pred, pred_err = net.predict_dataset(fits.getdata(contspac_file_name))
# unit conversion
nn_dist, nn_dist_err = fakemag_to_pc(pred[:, 0], corrected_K,
pred_err["total"][:, 0])
_, nn_dist_model_err = fakemag_to_pc(pred[:, 0], corrected_K,
pred_err["model"][:, 0])
nn_parallax, nn_parallax_err = fakemag_to_parallax(pred[:, 0], corrected_K,
pred_err["total"][:, 0])
_, nn_parallax_model_err = fakemag_to_parallax(pred[:, 0], corrected_K,
pred_err["model"][:, 0])
# remove astropy units
nn_dist = nn_dist.value
nn_dist_err = nn_dist_err.value
nn_dist_model_err = nn_dist_model_err.value
nn_parallax = nn_parallax.value
nn_parallax_err = nn_parallax_err.value
nn_parallax_model_err = nn_parallax_model_err.value
def test_astrometry_conversion(self):
# Example data of [Vega, Sirius]
absmag = np.array([0.582, 1.42])
mag = np.array([0.026, -1.46])
parallax = np.array([130.23, 379.21])
parallax_err = np.array([0.36, 1.58])
# Absmag related test
pc = absmag_to_pc(absmag, mag)
npt.assert_almost_equal(pc.value, 1000 / parallax, decimal=1)
npt.assert_almost_equal(mag_to_absmag(mag, parallax * u.mas),
absmag,
decimal=1)
self.assertEqual(
np.any(absmag_to_fakemag(MAGIC_NUMBER) == MAGIC_NUMBER), True)
absmag_test, absmag_err_test = mag_to_absmag(mag, parallax * u.mas,
parallax_err)
absmag_test_uniterr, absmag_err_test_uniterr = mag_to_absmag(
mag, parallax * u.mas, parallax_err / 1000 * u.arcsec)
absmag_test_arc, absmag_err_test_arc = mag_to_absmag(
mag, parallax / 1000 * u.arcsec, parallax_err / 1000)
absmag_test_unitless, absmag_err_test_unitless = mag_to_absmag(
mag, parallax, parallax_err)
# make sure unitless same as using astropy unit
npt.assert_almost_equal(absmag_test, absmag_test_unitless)
npt.assert_almost_equal(absmag_test, absmag_test_unitless)
npt.assert_almost_equal(absmag_err_test, absmag_err_test_uniterr)
# make sure astropy unit conversion works fine
npt.assert_almost_equal(absmag_test, absmag_test_arc)
npt.assert_almost_equal(absmag_err_test, absmag_err_test_arc)
# =================== Fakemag related test ===================#
# make sure these function did identity transform
npt.assert_almost_equal(fakemag_to_absmag(absmag_to_fakemag(absmag)),
absmag,
decimal=1)
# we can tests this after identity transformation confirmed
npt.assert_almost_equal(fakemag_to_pc(absmag_to_fakemag(absmag),
mag).value,
1000 / parallax,
decimal=1)
npt.assert_almost_equal(fakemag_to_pc(
mag_to_fakemag(mag, parallax * u.mas), mag).value,
1000 / parallax,
decimal=1)
fakemag_test, fakemag_err_test = mag_to_fakemag(
mag, parallax * u.mas, parallax_err)
fakemag_test_uniterr, fakemag_err_test_uniterr = mag_to_fakemag(
mag, parallax * u.mas, parallax_err / 1000 * u.arcsec)
fakemag_test_unitless, fakemag_err_test_unitless = mag_to_fakemag(
mag, parallax, parallax_err)
fakemag_test_arc, fakemag_err_test_arc = mag_to_fakemag(
mag, parallax / 1000 * u.arcsec, parallax_err / 1000)
pc_result, pc_result_err = fakemag_to_pc(fakemag_test, mag,
fakemag_err_test)
pc_result_arc, pc_result_err_arc = fakemag_to_pc(
fakemag_test_arc, mag, fakemag_err_test_arc)
# Analytically solution checkung
npt.assert_almost_equal(pc_result_err.value,
(parallax_err / parallax) * pc.value,
decimal=1)
# make sure unitless same as using astropy unit
npt.assert_almost_equal(fakemag_test, fakemag_test_unitless)
npt.assert_almost_equal(fakemag_err_test, fakemag_err_test_uniterr)
npt.assert_almost_equal(fakemag_err_test, fakemag_err_test_unitless)
# make sure astropy unit conversion works fine
npt.assert_almost_equal(fakemag_test, fakemag_test_arc)
npt.assert_almost_equal(fakemag_err_test, fakemag_err_test_arc)
npt.assert_almost_equal(pc_result.value, pc_result_arc.value)
npt.assert_almost_equal(pc_result_err.value, pc_result_err_arc.value)
# check gaia default dr
dr = gaia_default_dr()
self.assertEqual(dr, 1)
dr = gaia_default_dr(dr=3)
self.assertEqual(dr, 3)
apogeegaia_file = fits.getdata(gaia_rowmatch_f)
# add the offset we found for inv var weighting parallax
parallax = apogeegaia_file["parallax"] + 0.052
parallax_error = apogeegaia_file["parallax_error"]
# set negative parallax after constant offset correction to np.nan
parallax[(parallax < 0)] = np.nan
parallax_error[(parallax < 0)] = np.nan
# inference
net = load_folder(astronn_dist_model)
pred, pred_err = net.test(all_spec)
pred[:, 0][pred[:, 0] == 0.] = np.nan
# unit conversion
nn_dist, nn_dist_err = fakemag_to_pc(pred[:, 0], corrected_K,
pred_err['total'][:, 0])
_, nn_dist_model_err = fakemag_to_pc(pred[:, 0], corrected_K,
pred_err['model'][:, 0])
nn_parallax, nn_parallax_err = fakemag_to_parallax(pred[:, 0], corrected_K,
pred_err['total'][:, 0])
_, nn_parallax_model_err = fakemag_to_parallax(pred[:, 0], corrected_K,
pred_err['model'][:, 0])
# remove astropy units
nn_dist = nn_dist.value
nn_dist_err = nn_dist_err.value
nn_dist_model_err = nn_dist_model_err.value
nn_parallax = nn_parallax.value
nn_parallax_err = nn_parallax_err.value
nn_parallax_model_err = nn_parallax_model_err.value
def test_arXiv_1902_08634 (self):
"""
astroNN spectrophotometric distance
"""
from astroNN.apogee import visit_spectra, apogee_continuum
from astroNN.gaia import extinction_correction, fakemag_to_pc
from astropy.io import fits
# first model
models_url = ["https://github.com/henrysky/astroNN_gaia_dr2_paper/trunk/astroNN_no_offset_model",
"https://github.com/henrysky/astroNN_gaia_dr2_paper/trunk/astroNN_constant_model",
"https://github.com/henrysky/astroNN_gaia_dr2_paper/trunk/astroNN_multivariate_model"]
for model_url in models_url:
download_args = ["svn", "export", model_url]
res = subprocess.Popen(download_args, stdout=subprocess.PIPE)
output, _error = res.communicate()
if not _error:
pass
else:
raise ConnectionError(f"Error downloading the models {model_url}")
opened_fits = fits.open(visit_spectra(dr=14, location=4405, apogee='2M19060637+4717296'))
spectrum = opened_fits[1].data
spectrum_err = opened_fits[2].data
spectrum_bitmask = opened_fits[3].data
# using default continuum and bitmask values to continuum normalize
norm_spec, norm_spec_err = apogee_continuum(spectrum, spectrum_err,
bitmask=spectrum_bitmask, dr=14)
# correct for extinction
K = extinction_correction(opened_fits[0].header['K'], opened_fits[0].header['AKTARG'])
# ===========================================================================================#
# load neural net
neuralnet = load_folder('astroNN_no_offset_model')
# inference, if there are multiple visits, then you should use the globally
# weighted combined spectra (i.e. the second row)
pred, pred_err = neuralnet.test(norm_spec)
# convert prediction in fakemag to distance
pc, pc_error = fakemag_to_pc(pred[:, 0], K, pred_err['total'][:, 0])
# assert distance is close enough
# http://simbad.u-strasbg.fr/simbad/sim-id?mescat.distance=on&Ident=%406876647&Name=KIC+10196240&submit=display+selected+measurements#lab_meas
# no offset correction so further away
self.assertEqual(pc.value < 1250, True)
self.assertEqual(pc.value > 1100, True)
# ===========================================================================================#
# load neural net
neuralnet = load_folder('astroNN_constant_model')
# inference, if there are multiple visits, then you should use the globally
# weighted combined spectra (i.e. the second row)
pred, pred_err = neuralnet.test(np.hstack([norm_spec, np.zeros((norm_spec.shape[0], 4))]))
# convert prediction in fakemag to distance
pc, pc_error = fakemag_to_pc(pred[:, 0], K, pred_err['total'][:, 0])
# assert distance is close enough
# http://simbad.u-strasbg.fr/simbad/sim-id?mescat.distance=on&Ident=%406876647&Name=KIC+10196240&submit=display+selected+measurements#lab_meas
self.assertEqual(pc.value < 1150, True)
self.assertEqual(pc.value > 1000, True)
# ===========================================================================================#
# load neural net
neuralnet = load_folder('astroNN_multivariate_model')
# inference, if there are multiple visits, then you should use the globally
# weighted combined spectra (i.e. the second row)
pred, pred_err = neuralnet.test(np.hstack([norm_spec, np.zeros((norm_spec.shape[0], 4))]))
# convert prediction in fakemag to distance
pc, pc_error = fakemag_to_pc(pred[:, 0], K, pred_err['total'][:, 0])
# assert distance is close enough
# http://simbad.u-strasbg.fr/simbad/sim-id?mescat.distance=on&Ident=%406876647&Name=KIC+10196240&submit=display+selected+measurements#lab_meas
self.assertEqual(pc.value < 1150, True)
self.assertEqual(pc.value > 1000, True)
