def test_rng_seed_equal():
N = 64
x = normal(0, 1, N)
seed = randint(1 << 30)
imfs1 = eemd(x, rng_seed=seed)
imfs2 = eemd(x, rng_seed=seed)
assert_allclose(imfs1, imfs2)
def test_ones():
N = 64
x = [1] * N
imfs = eemd(x, ensemble_size=100)
for n in range(imfs.shape[0] - 1):
imf = imfs[n, :]
assert all(abs(imf) < 1e-9)
assert_allclose(imfs[-1, :], x)
def test_extract_residual():
N = 100
t = linspace(1, 10, num=N)
x = t**2
xn = x + normal(0, 0.5, N)
imfs = eemd(xn)
# the residual should be approximately equal to the signal without the
# added noise, at least away from the ends
residual = imfs[-1, :]
print(abs(residual - x)[10:-10])
assert_allclose(residual[10:-10], x[10:-10], rtol=0.1, atol=1)
def test_zeros():
x = zeros(64)
imfs = eemd(x, ensemble_size=10)
# the zero signal has zero standard deviation so no noise should be added
assert all(imfs == 0)
def test_invalid_arguments8():
x = []
eemd(x, num_imfs=-5)
def test_invalid_arguments6():
x = []
eemd(x, num_imfs="Lots")
def test_invalid_arguments4():
x = []
eemd(x, num_siftings=-3)
def test_invalid_arguments3():
x = []
eemd(x, num_siftings=0, S_number=0)
def test_invalid_arguments2():
x = []
eemd(x, noise_strength=-2)
def test_wrong_dims():
x = zeros((2, 2))
eemd(x)
def test_bogus2():
x = eemd(7)
def test_bogus1():
x = eemd("I am a banana")
def test_num_imfs_output_size():
N = 64
x = normal(0, 1, N)
imfs = eemd(x, num_imfs=3)
assert imfs.shape[0] == 3
def test_num_imfs():
N = 64
x = normal(0, 1, N)
imfs1 = eemd(x, num_imfs=3, num_siftings=10, rng_seed=1234)
imfs2 = eemd(x, num_imfs=4, num_siftings=10, rng_seed=1234)
assert_allclose(imfs1[:2, :], imfs2[:2, :])
def test_rng_seed_nonequal():
N = 64
x = normal(0, 1, N)
imfs1 = eemd(x, rng_seed=3141)
imfs2 = eemd(x, rng_seed=5926)
assert not allclose(imfs1, imfs2)
def test_invalid_arguments1():
x = []
eemd(x, ensemble_size=0)
if __name__ == "__main__":
# os settings
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
os.environ['KMP_DUPLICATE_LIB_OK'] = 'TRUE'
# load data from csv file
data = pd.read_csv('data.csv', index_col='date', parse_dates=['date'])
# normalize data
data_max = data.max()
data_min = data.min()
data = (data - data_min) / (data_max - data_min)
# get imfs via eemd
imfs = pyeemd.eemd(data.values.reshape(-1))
# plot imfs
# i = 1
# plt.figure(2)
# for imf in imfs:
# plt.subplot(len(imfs), 1, i)
# plt.plot(imf)
# i += 1
# plt.show()
timestep = 10
train_size = int(len(data) * 0.8)
# i = 1
imfs_prediction = []
import numpy as np
from numpy import pi
# An example signal with a lower frequency sinusoid modulated by an
# intermittent higher-frequency sinusoid
x = np.linspace(0, 2 * pi, num=500)
signal = np.sin(4 * x)
intermittent = 0.1 * np.sin(80 * x)
y = signal * (1 + np.select([signal > 0.7], [intermittent]))
figure()
title("Original signal")
plot(x, y)
# Decompose with EEMD
imfs = eemd(y, num_siftings=10)
# Plot high-frequency IMFs (1-3) and the rest separately. This illustrates how EEMD can extract the intermittent signal.
highfreq_sum = np.sum([imfs[i] for i in range(0, 3)], axis=0)
lowfreq_sum = np.sum([imfs[i] for i in range(3, imfs.shape[0])], axis=0)
figure()
title("Sum of IMFs 1 to 3")
plot(x, highfreq_sum)
figure()
title("Sum of rest of IMFs")
plot(x, lowfreq_sum)
show()
