You can not select more than 25 topics
			Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
		
		
		
		
			
				
					67 lines
				
				1.6 KiB
			
		
		
			
		
	
	
					67 lines
				
				1.6 KiB
			| 
											6 years ago
										 | import bisect
 | ||
|  | import numpy as np
 | ||
|  | from scipy.interpolate import interp1d
 | ||
|  | 
 | ||
|  | 
 | ||
|  | def deep_interp_0_fast(dx, x, y):
 | ||
|  |   FIX = False
 | ||
|  |   if len(y.shape) == 1:
 | ||
|  |     y = y[:, None]
 | ||
|  |     FIX = True
 | ||
|  |   ret = np.zeros((dx.shape[0], y.shape[1]))
 | ||
|  |   index = list(x)
 | ||
|  |   for i in range(dx.shape[0]):
 | ||
|  |     idx = bisect.bisect_left(index, dx[i])
 | ||
|  |     if idx == x.shape[0]:
 | ||
|  |       idx = x.shape[0] - 1
 | ||
|  |     ret[i] = y[idx]
 | ||
|  | 
 | ||
|  |   if FIX:
 | ||
|  |     return ret[:, 0]
 | ||
|  |   else:
 | ||
|  |     return ret
 | ||
|  | 
 | ||
|  | 
 | ||
|  | def running_mean(x, N):
 | ||
|  |   cumsum = np.cumsum(np.insert(x, [0]*(int(N/2)) + [-1]*(N-int(N/2)), [x[0]]*int(N/2) + [x[-1]]*(N-int(N/2))))
 | ||
|  |   return (cumsum[N:] - cumsum[:-N]) / N
 | ||
|  | 
 | ||
|  | 
 | ||
|  | def deep_interp_np(x, xp, fp):
 | ||
|  |   x = np.atleast_1d(x)
 | ||
|  |   xp = np.array(xp)
 | ||
|  |   if len(xp) < 2:
 | ||
|  |     return np.repeat(fp, len(x), axis=0)
 | ||
|  |   if min(np.diff(xp)) < 0:
 | ||
|  |     raise RuntimeError('Bad x array for interpolation')
 | ||
|  |   j = np.searchsorted(xp, x) - 1
 | ||
|  |   j = np.clip(j, 0, len(xp)-2)
 | ||
|  |   d = np.divide(x - xp[j], xp[j + 1] - xp[j], out=np.ones_like(x, dtype=np.float64), where=xp[j + 1] - xp[j] != 0)
 | ||
|  |   vals_interp = (fp[j].T*(1 - d)).T + (fp[j + 1].T*d).T
 | ||
|  |   if len(vals_interp) == 1:
 | ||
|  |     return vals_interp[0]
 | ||
|  |   else:
 | ||
|  |     return vals_interp
 | ||
|  | 
 | ||
|  | 
 | ||
|  | def clipping_deep_interp(x, xp, fp):
 | ||
|  |   if len(xp) < 2:
 | ||
|  |     return deep_interp_np(x, xp, fp)
 | ||
|  |   bad_idx = np.where(np.diff(xp) < 0)[0]
 | ||
|  |   if len(bad_idx) > 0:
 | ||
|  |     if bad_idx[0] ==1:
 | ||
|  |       return np.zeros([] + list(fp.shape[1:]))
 | ||
|  |     return deep_interp_np(x, xp[:bad_idx[0]], fp[:bad_idx[0]])
 | ||
|  |   else:
 | ||
|  |     return deep_interp_np(x, xp, fp)
 | ||
|  | 
 | ||
|  | 
 | ||
|  | def deep_interp(dx, x, y, kind="slinear"):
 | ||
|  |   return interp1d(
 | ||
|  |     x, y,
 | ||
|  |     axis=0,
 | ||
|  |     kind=kind,
 | ||
|  |     bounds_error=False,
 | ||
|  |     fill_value="extrapolate",
 | ||
|  |     assume_sorted=True)(dx)
 |