def airpls_baseline(intensities, smoothness_param=100, max_iters=10,
conv_thresh=0.001, verbose=False):
'''
Baseline corr. using adaptive iteratively reweighted penalized least squares.
Also known as airPLS, 2010.
http://pubs.rsc.org/EN/content/articlehtml/2010/an/b922045c
https://code.google.com/p/airpls/
https://airpls.googlecode.com/svn/trunk/airPLS.py
'''
smoother = WhittakerSmoother(intensities, smoothness_param)
total_intensity = np.abs(intensities).sum()
w = np.ones(intensities.shape[0])
for i in range(1, max_iters+1):
baseline = smoother.smooth(w)
# Compute error (sum of distances below the baseline).
corrected = intensities - baseline
mask = corrected < 0
baseline_error = -corrected[mask]
total_error = baseline_error.sum()
# Check convergence as a fraction of total intensity.
conv = total_error / total_intensity
if verbose:
print(i, conv)
if conv < conv_thresh:
break
# Set peak weights to zero.
w[~mask] = 0
# Set baseline weights.
baseline_error /= total_error
w[mask] = np.exp(i*baseline_error)
w[0] = np.exp(i*baseline_error.min())
w[-1] = w[0]
else:
print('airPLS did not converge in %d iterations' % max_iters)
return baseline
def als_baseline(intensities, asymmetry_param=0.05, smoothness_param=1e6,
max_iters=10, conv_thresh=1e-5, verbose=False):
'''Perform asymmetric least squares baseline removal.
* http://www.science.uva.nl/~hboelens/publications/draftpub/Eilers_2005.pdf
smoothness_param: Relative importance of smoothness of the predicted response.
asymmetry_param (p): if y > z, w = p, otherwise w = 1-p.
Setting p=1 is effectively a hinge loss.
'''
smoother = WhittakerSmoother(intensities, smoothness_param, deriv_order=2)
# Rename p for concision.
p = asymmetry_param
# Initialize weights.
w = np.ones(intensities.shape[0])
for i in range(max_iters):
z = smoother.smooth(w)
mask = intensities > z
new_w = p*mask + (1-p)*(~mask)
conv = np.linalg.norm(new_w - w)
if verbose:
print(i+1), conv
if conv < conv_thresh:
break
w = new_w
else:
print('ALS did not converge in %d iterations' % max_iters)
return z
def fabc_baseline(intensities, dilation_param=50, smoothness_param=1e3): '''Fully Automatic Baseline Correction, by Carlos Cobas (2006). http://www.sciencedirect.com/science/article/pii/S1090780706002266 ''' cwt = scipy.signal.cwt(intensities, scipy.signal.ricker, (dilation_param,)) dY = cwt.ravel()**2 is_baseline = iterative_threshold(dY) is_baseline[0] = True is_baseline[-1] = True smoother = WhittakerSmoother(intensities, smoothness_param, deriv_order=1) return smoother.smooth(is_baseline)
