Zeel's picture
t
9af7384
import numpy as np
from scipy.interpolate import bisplrep, bisplev
from ..base import Base
from ..utils import find_closest
class Spline(Base):
"""
Class to use a bivariate B-spline to interpolate values.
https://docs.scipy.org/doc/scipy-0.18.1/reference/generated/scipy.interpolate.bisplrep.html#scipy.interpolate.bisplrep
Parameters
----------
kx, ky: int, int, optional
The degrees of the spline (1 <= kx, ky <= 5).
Third order (kx=ky=3) is recommended.
s : float, optional
A non-negative smoothing factor. If weights correspond
to the inverse of the standard-deviation of the errors
in z, then a good s-value should be found in the
range `(m-sqrt(2*m),m+sqrt(2*m))` where `m=len(x)`.
"""
def __init__(
self,
kx=3,
ky=3,
s=None,
resolution="standard",
coordinate_type="Euclidean",
):
super().__init__(resolution, coordinate_type)
self.kx = kx
self.ky = ky
self.s = s
def _fit(self, X, y):
"""The function call to fit the spline model on the given data.
This function is not supposed to be called directly.
"""
# fitting the curve
# bisplrep returns details of the fitted curve
# read bisplrep docs for more info about it's return values.
self.tck = bisplrep(
X[:, 0], X[:, 1], y, kx=self.kx, ky=self.ky, s=self.s
)
return self
def _predict_grid(self, x1lim, x2lim):
"""The function to predict grid interpolation using the BSpline.
This function is not supposed to be called directly.
"""
# getting the boundaries for interpolation
x1min, x1max = x1lim
x2min, x2max = x2lim
# interpolating over the grid
# TODO Relook here, we might expect the result to be transpose
return bisplev(
np.linspace(x1min, x1max, self.resolution),
np.linspace(x2min, x2max, self.resolution),
self.tck,
)
def _predict(self, X):
"""The function to predict using the BSpline interpolation.
This function is not supposed to be called directly.
"""
results = []
for ix in range(X.shape[0]):
interpolated_y = bisplev(
X[ix, 0], X[ix, 1], self.tck
).item() # one value returned
results.append(interpolated_y)
return np.array(results)
# # form a grid
# x1 = np.linspace(self.x1min_d, self.x1max_d, self.resolution),
# x2 = np.linspace(self.x2min_d, self.x2max_d, self.resolution),
# X1, X2 = np.meshgrid(x1, x2)
# # be default run grid interpolation on the whole train data
# interpolated_grid = bisplev(
# x1, x2,
# self.tck,
# )
# # find the closest points on the interpolated grid
# ix = find_closest(grid=(X1, X2), X)
# return interpolated_grid[ix] # TODO this can be wrong, must depend on