In [15]:

    

%load_ext autoreload
%autoreload 2


    

In [14]:

    

# NBVAL_IGNORE_OUTPUT

import numpy as np
import matplotlib.pyplot as plt
from toolz.curried import pipe, compose, map
from fmks.bases.primitive import primitive_basis
from fmks.localization import fit, predict, coeff_to_real
from typing import Tuple
import scipy.ndimage
from fmks.func import curry
import dask.array as da


    

In [3]:

    

def filter(x):
    return np.exp(-abs(x)) * np.cos(2 * np.pi * x)


    

In [12]:

    

# NBVAL_IGNORE_OUTPUT

def fplot(x, y):
    plt.plot(x, y, color='#1a9850')
    plt.show()

io = pipe(
        np.linspace(-10.0, 10.0, 1000),
        lambda x: fplot(x, filter(x))
)


    

In [5]:

    

def get_x_data(n_sample, n_space):
    np.random.seed(201)
    return np.random.random((n_sample, n_space))


    

In [6]:

    

@curry
def convolve(x_sample):
    return scipy.ndimage.convolve(x_sample,
                                  filter(np.linspace(-10.0, 10.0, len(x_sample))),
                                  mode="wrap")


    

In [7]:

    

get_y_data = compose(np.array, list, map(convolve))
get_xy_data = lambda x: (x, get_y_data(x))
    
data = get_xy_data(get_x_data(50, 101))


    

In [8]:

    

print(data[0].shape, data[1].shape)


    

    




(50, 101) (50, 101)

In [9]:

    

def fit_predict(x_data, y_data):
    basis = primitive_basis(n_state=2)
    return pipe(
        fit(x_data, y_data, basis),
        lambda coeff: (predict(x_data, coeff, basis), coeff)
    )

x, y = data
y_pred, coeff = fit_predict(da.from_array(x, chunks=(10, 101)),
                            da.from_array(y, chunks=(10, 101)))
print(y_pred[0, :4])
print(y[0, :4])
print(coeff.shape)


    

    




[-0.41059557  0.20004566  0.61200171  0.5878077 ]
[-0.41059557  0.20004566  0.61200171  0.5878077 ]
(51, 2)

In [10]:

    

# NBVAL_IGNORE_OUTPUT

x = np.linspace(-10.0, 10.0, 101)
rcoeff = coeff_to_real(coeff, (101,))
print(x.shape)
print(rcoeff.shape)
plt.plot(x, filter(x), label=r'$F$', color='#1a9850')
plt.plot(x, -rcoeff[:, 0] + rcoeff[:, 1], 
         'k--', label=r'$\alpha$')
l = plt.legend()
plt.show()


    

    




(101,)
(101, 2)






    

In [11]:

    

print(y_pred[0, :4])


    

    




[-0.41059557  0.20004566  0.61200171  0.5878077 ]

In [ ]: