Error LinAlgError: SVD did not converge in Linear Least Squares

[abearab]

When I try running blitz.gsea function in jupyter for multiple tests in a for loop, I get this error message.

LinAlgError: SVD did not converge in Linear Least Squares

---

full message

python3.7/site-packages/blitzgsea/__init__.py in gsea(signature, library, permutations, anchors, min_size, max_size, processes, plotting, verbose, symmetric, signature_cache, seed)
    239         f_alpha_pos, f_beta_pos, f_pos_ratio, ks_pos, ks_neg = blitzgsea_signature_anchors[sig_hash]
    240     else:
--> 241         f_alpha_pos, f_beta_pos, f_pos_ratio, ks_pos, ks_neg = estimate_parameters(signature, abs_signature, signature_map, library, permutations=permutations, calibration_anchors=anchors, processes=processes, symmetric=symmetric, plotting=plotting, verbose=verbose, seed=seed)
    242         blitzgsea_signature_anchors[sig_hash] = (f_alpha_pos, f_beta_pos, f_pos_ratio, ks_pos, ks_neg)
    243 

python3.7/site-packages/blitzgsea/__init__.py in estimate_parameters(signature, abs_signature, signature_map, library, permutations, symmetric, calibration_anchors, plotting, processes, verbose, seed)
    120 
    121     f_alpha_pos = loess_interpolation(x, alpha_pos)
--> 122     f_beta_pos = loess_interpolation(x, beta_pos, frac=0.2)
    123 
    124     f_alpha_neg = loess_interpolation(x, alpha_neg)

python3.7/site-packages/blitzgsea/__init__.py in loess_interpolation(x, y, frac)
     76 def loess_interpolation(x, y, frac=0.5):
     77     yl = np.array(y)
---> 78     xout, yout, wout = loess_1d(x, yl, frac=frac)
     79     return interpolate.interp1d(xout, yout)
     80 

python3.7/site-packages/loess/loess_1d.py in loess_1d(x, y, xnew, degree, frac, npoints, rotate, sigy)
    274             biweights = (1 - uu)**2
    275             tot_weights = dist_weights*biweights
--> 276             poly = polyfit1d(x[w], y[w], degree, tot_weights)
    277             yfit = poly.yfit
    278             badOld = bad

python3.7/site-packages/loess/loess_1d.py in __init__(self, x, y, degree, weights)
     81         a = x[:, None]**np.arange(degree + 1)
     82         self.degree = degree
---> 83         self.coeff = np.linalg.lstsq(a*sqw[:, None], y*sqw, rcond=None)[0]
     84         self.yfit = a @ self.coeff
     85 

<__array_function__ internals> in lstsq(*args, **kwargs)

python3.7/site-packages/numpy/linalg/linalg.py in lstsq(a, b, rcond)
   2304         # lapack can't handle n_rhs = 0 - so allocate the array one larger in that axis
   2305         b = zeros(b.shape[:-2] + (m, n_rhs + 1), dtype=b.dtype)
-> 2306     x, resids, rank, s = gufunc(a, b, rcond, signature=signature, extobj=extobj)
   2307     if m == 0:
   2308         x[...] = 0

python3.7/site-packages/numpy/linalg/linalg.py in _raise_linalgerror_lstsq(err, flag)
     98 
     99 def _raise_linalgerror_lstsq(err, flag):
--> 100     raise LinAlgError("SVD did not converge in Linear Least Squares")
    101 
    102 def get_linalg_error_extobj(callback):

[abearab]

it seems more like a @numpy / @scipy issue – jaakkopasanen/AutoEq#256

I had:

scipy.__version__
'1.7.3'
np.__version__
'1.19.1'

[abearab]

Okay! I need to avoid small sets in the library input for running gsea function.