A simple tutorial to Stellar LAbel Machine (SLAM)¶
Bo Zhang (mailto:bozhang@nao.cas.cn), Created on Thu Jan 19 15:48:12 2017
In [2]:
import numpy as np
import matplotlib.pyplot as plt
import os
from slam.slam import Slam
from slam.diagnostic import compare_labels
from slam.apogee import aspcapStar_read, apStar_read
from slam.normalization import normalize_spectrum, normalize_spectra_block
from slam.binning import interp_pchip, interp_linear, interp_nearest
from astropy.table import Table, Column
from joblib import Parallel, delayed, dump, load
os.chdir("/home/cham/PycharmProjects/slam/doc/example_DR10/Data")
In [3]:
""" load catalog """
t = Table.read('../reference_labels.csv')
label3 = np.array(t['Teff_{corr}', 'logg_{corr}', '[M/H]_{corr}'].to_pandas())
In [4]:
""" define functions """
def apStar_read_interp(fp, wave_interp, i_use=0):
spec = apStar_read(fp, full=True, meta=True, verbose=False)
wave_rest = spec['wave']#/(1+rv/299792.458)
if spec.meta['NVISITS'] > 1:
flux_interp = interp_linear(wave_rest, spec['flux'][:, i_use], wave_interp, 0)
ivar_interp = (1./interp_linear(wave_rest, spec['flux_err'][:, i_use], wave_interp, 1E10))**2
mask_interp = interp_nearest(wave_rest, spec['mask'][:, i_use], wave_interp, 1)
else:
flux_interp = interp_linear(wave_rest, spec['flux'], wave_interp, 0)
ivar_interp = (1./interp_linear(wave_rest, spec['flux_err'], wave_interp, 1E10))**2
mask_interp = interp_nearest(wave_rest, spec['mask'], wave_interp, 1)
return flux_interp, ivar_interp, mask_interp
def apStar_read_block(fps, wave_interp, n_jobs=1, verbose=False):
r = Parallel(n_jobs=n_jobs, verbose=verbose)(
delayed(apStar_read_interp)(fp,wave_interp,0) for fp in fps)
flux_block = np.array([_[0] for _ in r])
ivar_block = np.array([_[1] for _ in r])
mask_block = np.array([_[2] for _ in r])
return flux_block, ivar_block, mask_block
def aspcapStar_read_interp(fp, wave_interp, i_use=0):
spec = aspcapStar_read(fp, meta=True)
return spec['flux'], 1./spec['flux_err'].data**2.
def aspcapStar_read_block(fps, n_jobs=1, verbose=False):
r = Parallel(n_jobs=n_jobs, verbose=verbose)(
delayed(aspcapStar_read_interp)(fp,wave_interp,0) for fp in fps)
flux_block = np.array([_[0] for _ in r])
ivar_block = np.array([_[1] for _ in r])
ivar_block = np.where(np.isfinite(ivar_block), ivar_block, np.zeros_like(ivar_block))
return flux_block, ivar_block
In [5]:
""" define wavelength grid according to the first spectrum """
spec = aspcapStar_read(t['ID'][0], True, True)
wave_interp = spec['wave'].data
""" read all spectra """
flux_block, ivar_block = aspcapStar_read_block(t['ID'].data, n_jobs=-1)
""" normalization """
r1 = normalize_spectra_block(
wave_interp, flux_block, (15200., 17000.), 30., p=(1E-8, 1E-7), q=0.7,
eps=1E-19, ivar_block=flux_block > 0, rsv_frac=1., n_jobs=10, verbose=5)
flux_norm, flux_cont = r1
ivar_norm = flux_cont**2*ivar_block
In [6]:
""" training """
k = Slam(wave_interp, flux_norm, ivar_norm, label3)
# train: using simple user-defined hyper-parameters
# it takes ~2 min using 32 cores
k.train_pixels(method='simple', n_jobs=24, verbose=5, C=2.0, epsilon=0.1, gamma=1.)
In [7]:
""" test """
# here simply to test SLAM on training sample
slc = slice(0, flux_norm.shape[0], 20)
# 1. derive stellar parameters using template matching
# it takes a few minutes
label_init = k.predict_labels_quick(flux_norm[slc, :], ivar_norm[slc, :], n_jobs=24)
# 2. re-fine the initial guess
# it takes an hour ...
label_refined = k.predict_labels_multi(label_init, flux_norm[slc, :], ivar_norm[slc, :], n_jobs=24)
In [10]:
""" plot results """
fig = compare_labels(label3[slc, :], label_refined, 'ASPCAP', 'SLAM')
In [11]:
fig
Out[11]:
In [ ]: