Develop Reference

Speed up for loop in python

I have a for loop as follows:
import MDAnalysis as mda
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from tqdm import tqdm as tq
import MDAnalysis.analysis.pca as pca
import random
def PCA_projection(pdb,dcd,atomgroup):
u = mda.Universe(pdb,dcd)
PSF_pca = pca.PCA(u, select=atomgroup)
PSF_pca.run(verbose=True)
n_pcs = np.where(PSF_pca.results.cumulated_variance > 0.95)[0][0]
atomgroup = u.select_atoms(atomgroup)
pca_space = PSF_pca.transform(atomgroup, n_components=n_pcs)
PC1_proj = [pca_space[i][0] for i in range(len(pca_space))]
PC2_proj = [pca_space[i][1] for i in range(len(pca_space))]
return PC1_proj, PC2_proj
def Read_bias_potential(bias_potential):
Bias_potential = pd.read_csv(bias_potential)
Bias_potential = Bias_potential['En-User']
Bias_potential = Bias_potential.values.tolist()
W = [math.exp((-1 * i) / (0.001987*300)) for i in Bias_potential]
return W
def Bin(PC1_prj, PC2_prj, frame_num, min_br1, max_br1, min_br2, max_br2, bin_num, W):
#import pdb;pdb.set_trace()
data1 = PC1_prj[0:frame_num]
bins1 = np.linspace(min_br1, max_br1, bin_num)
bins1 = np.round(bins1,2)
digitized1 = np.digitize(data1, bins1)
binc1 = np.arange(min_br1 + (max_br1 - min_br1)/2*bin_num,
max_br1 + (max_br1 - min_br1)/2*bin_num, (max_br1 - min_br1)/bin_num, dtype = float)
binc1 = np.around(binc1,3)
data2 = PC2_prj[0:frame_num]
bins2 = np.linspace(min_br2, max_br2, bin_num)
bins2 = np.round(bins2,2)
digitized2 = np.digitize(data2, bins2)
binc2 = np.arange(min_br2 + (max_br2 - min_br2)/2*bin_num, max_br2 + (max_br2 - min_br2)/2*bin_num, (max_br2 - min_br2)/bin_num, dtype = float)
binc2 = np.around(binc2,3)
w_array = np.zeros((bin_num,bin_num))
for j in range(frame_num):
w_array[digitized1[j]][digitized2[j]] += (W[digitized1[j]] + W[digitized2[j]])
for m in range(bin_num):
for n in range(bin_num):
if w_array[m][n] == 0:
w_array[m][n] = 1e-100
return w_array, binc1, binc2
def gaussian(Sj1,Slj1,Sj2,Slj2,count):
sigma1 = 0.5
sigma2 = 0.5
Kb = 0.001987204
T = 300
h0 = 0.0001
g = 0
C1 = 0
C2 = 0
for i in range((np.where(Slj2 == Sj2)[0][0] - 5),(np.where(Slj2 == Sj2)[0][0] + 6)):
if i < 0:
C2 = i + 1000
elif i > 999:
C2 = i - 1000
else:
C2 = i
for j in range((np.where(Slj1 == Sj1)[0][0] - 5),(np.where(Slj2 == Sj2)[0][0] + 6)):
if j < 0:
C1 = j + 1000
elif j > 999:
C1 = j -1000
else:
C1 = j
g = g + count[C2,C1] * h0 * np.exp( (-(Sj1 - Slj1[C1]) ** 2 / (2 * sigma1 ** 2)) + (-(Sj2 - Slj2[C2]) ** 2 / (2 * sigma2 ** 2)) )
return np.exp(-g / (Kb * T))
def resampling(binc1, binc2, w_array):
# import pdb;pdb.set_trace()
l =1000
F = np.zeros((l,l))
count = np.zeros((l,l))
Wn = w_array
for i in tq(range(10000000)):
SK1 = random.choice(binc1)
SK2 = random.choice(binc2)
SL1 = random.choice(binc1)
SL2 = random.choice(binc2)
while SK1 == SL1:
SL1 = random.choice(binc1)
while SK2 == SL2:
SL2 = random.choice(binc2)
F[np.where(binc2 == SK2)[0][0]][np.where(binc1 == SK1)[0][0]] = gaussian(SK1,binc1,SK2,binc2,count)
F[np.where(binc2 == SK2)[0][0]][np.where(binc1 == SK1)[0][0]] = gaussian(SL1,binc1,SL2,binc2,count)
W_SK = Wn[np.where(binc2 == SK2)[0][0]][np.where(binc1 == SK1)[0][0]] * F[np.where(binc2 == SK2)[0][0]][np.where(binc1 == SK1)[0][0]]
W_SL = Wn[np.where(binc2 == SL2)[0][0]][np.where(binc1 == SL1)[0][0]] * F[np.where(binc2 == SL2)[0][0]][np.where(binc1 == SL1)[0][0]]
if W_SK <= W_SL:
SK1 = SL1
SK2 = SL2
else:
a = random.random()
if W_SL/W_SK >= a:
SK1 = SL1
SK2 = SL2
else:
SK1 = SK1
SK2 = SK2
#print('SK =',SK)
count[np.where(binc2 == SK2)[0][0]][np.where(binc1 == SK1)[0][0]] += 1
return F
where binc1 and binc2 are two np.arrays, gaussian is a gaussian fxn I defined, is there anyway I can speed up this for loop? Now 1000000 steps takes approximately 50 mins. I am thinking about using pytorch but I got no idea on how to do it. Any suggestions would be helpful!
Thanks
I tried to use pytorch, like put all the variables on gpu but it only does worse.

Am trying to append a matrix to a numpy array from a 3d list but It only append first row and when I try to append whole matrix It returns an error

I have a certain number of matrix in a list and when I try to append a matrix to the NumPy array It only appends the specified row and when I try to edit the code to append the whole matrix It keep returning the following error:
Traceback (most recent call last):
File "so.py", line 129, in <module>
parents = selection(cal_fitness(t1, t2, t3, matlist, 300, arr), 2, matlist) #
File "so.py", line 124, in selection
parents[i, :] = population[max_fitness_idx[0][0]]
ValueError: could not broadcast input array from shape (3,5) into shape (5,)
The error related function:
def selection(fitness, num_parents, population):
fitness = list(fitness)
parents = numpy.empty((num_parents, len(population)))
for i in range(num_parents):
max_fitness_idx = numpy.where(fitness == numpy.max(fitness))
#parents[i, :] = population[max_fitness_idx[0][0]][2]
parents[i, :] = population[max_fitness_idx[0][0]]
fitness[max_fitness_idx[0][0]] = -999999
return parents.astype(int)
parents = selection(cal_fitness(t1, t2, t3, matlist, 300, arr), 2, matlist)
print(parents)
The full code:
import numpy
import random
import pandas
wsn = numpy.arange(1, 6)
taskn = 3
t1 = numpy.random.randint(30, 200, size=len(wsn))
t2 = numpy.random.randint(30, 200, size=len(wsn))
t3 = numpy.random.randint(30, 200, size=len(wsn))
# print('\nGenerated Data:\t\n\nNumber \t Task 1 \t Task 2 \t Task 3\n')
ni = min(len(t1), len(t2), len(t3))
# for i in range(ni):
# print('\t {0} \t {1} \t {2} \t\t {3}\n'.format(wsn[i], t1[i], t2[i], t3[i]))
# print('\n\n')
qmin = 50
qmax = 140
for i in range(len(t1)):
if t1[i] <= qmin or t1[i] >= qmax:
# t1=numpy.delete(t1,i)
t1[i] = 0
for i in range(len(t2)):
if t2[i] <= qmin or t2[i] >= qmax:
# t2=numpy.delete(t2,i)
t2[i] = 0
for i in range(len(t3)):
if t3[i] <= qmin or t3[i] >= qmax:
# t3=numpy.delete(t3,i)
t3[i] = 0
i = 0
m = max(len(t1), len(t2), len(t3))
if t1[i] == 0 and t2[i] == 0 and t3[i] == 0:
t1 = numpy.delete(t1, i)
t2 = numpy.delete(t2, i)
t3 = numpy.delete(t3, i)
i += 1
solperpop = len(wsn)
gen = 20
j = 0
pop_size = (taskn, solperpop)
# print('population size: {}'.format(pop_size))
# for j in range(ni):
# pop_size=list(solperpop,taskn)
matlist = list()
# print('\n\n')
i = 0
k = 0
nbrofindv = 5
arr = []
for i in range(nbrofindv):
init_pop = numpy.zeros(pop_size, dtype=int)
init_pop = init_pop.astype(int)
k = 0
l = 0
for k in range(taskn):
l = random.randrange(solperpop - 1)
init_pop[k][l] = 1
arr.append(l)
matlist.append(init_pop)
pandas.set_option('display.max_columns', None)
pandas.set_option('display.width', None)
zipped = pandas.DataFrame(list(zip(*matlist)))
# , columns=['Individual 1', 'Individual 2', 'Individual 3', 'Individual 4', 'Individual 5'])
print(zipped)
print('\n\n')
i = 0
for i in range(len(wsn)):
if t1[i] == 0:
if init_pop[0][i] != 0:
init_pop[0][i] == 0
if t2[i] == 0:
if init_pop[1][i] != 0:
init_pop[1][i] == 0
if t3[i] == 0:
if init_pop[2][i] != 0:
init_pop[2][i] == 0
def cal_fitness(task1, task2, task3, matix, mmax, array):
fitness = numpy.empty(len(matix))
S1 = numpy.empty(len(matix), dtype=int)
z = 0
for i in range(len(matix)):
S1[i] = task1[array[0 + z]] + task2[array[1 + z]] + task3[array[2 + z]]
z += 3
if S1[i] <= mmax:
fitness[i] = S1[i]
else:
fitness[i] = 0
return fitness.astype(int)
fitness = cal_fitness(t1, t2, t3, matlist, 300, arr)
def selection(fitness, num_parents, population):
fitness = list(fitness)
parents = numpy.empty((num_parents, len(population)))
for i in range(num_parents):
max_fitness_idx = numpy.where(fitness == numpy.max(fitness))
#parents[i, :] = population[max_fitness_idx[0][0]][2]
parents[i, :] = population[max_fitness_idx[0][0]]
fitness[max_fitness_idx[0][0]] = -999999
return parents.astype(int)
parents = selection(cal_fitness(t1, t2, t3, matlist, 300, arr), 2, matlist)
print(parents)
print('\n\n')
def crossover(parents, num_offsprings):
offsprings = numpy.empty((num_offsprings, parents.shape[1]))
crossover_point = int(parents.shape[1] / 2)
crossover_rate = 0.5
i = 0
while parents.shape[0] < num_offsprings:
parent1_index = i % parents.shape[0]
parent2_index = (i + 1) % parents.shape[0]
x = random.random()
if x > crossover_rate:
continue
parent1_index = i % parents.shape[0]
parent2_index = (i + 1) % parents.shape[0]
offsprings[i, 0:crossover_point] = parents[parent1_index, 0:crossover_point]
offsprings[i, crossover_point:] = parents[parent2_index, crossover_point:]
i += 1 # <== modified
return offsprings # <== modified
print(crossover(parents, 2)) # <== modified

This issue is related to dimensions that specified for parents, which is 2d now so must be modified to 3d. So, IIUC, you can achieve this aim by:
def selection(fitness, num_parents, population):
fitness = list(fitness)
# parents shape is modified (a new axis with the needed length is added)
parents = numpy.empty((num_parents, len(population[0]), len(population[0][0]))) # <==
for i in range(num_parents):
max_fitness_idx = numpy.where(fitness == numpy.max(fitness))
parents[i, :] = population[max_fitness_idx[0][0]]
fitness[max_fitness_idx[0][0]] = -999999
return parents.astype(int)

How can I apply image write file path?

i = 2 #int
cv2.imwrite([r'C:\Users\Desktop\result (' + str(i) + ').png'], result) #result is 16bit image
I want to save image under the name 'result (2).png'
Because, i is stuck in 'for loop'.
However, the code above causes an error.
Please help me.
add)
## Flat Field Correction (FFC) ##
import numpy as np
import cv2
import matplotlib.pyplot as plt
import numba as nb
import multiprocessing as multi
import parmap
import time
start = time.time()
B = cv2.imread(r'D:\remedi\Exercise\Xray\Offset.png', -1) # offset image
for i in range(2,3):
org_I = cv2.imread(r'D:\remedi\Exercise\Xray\objects\object (' + str(i) + ').png', -1) # original image
w = cv2.imread(r'D:\remedi\Exercise\Xray\white\white (' + str(i) + ').png', -1) # white image
## dead & bad pixel correction
corrected_w = w.copy()
corrected_org_I = org_I.copy()
c = np.mean(corrected_w)
p = np.abs(corrected_w - c)
sens = 0.7
[num_y, num_x] = np.where((p < c*sens) | (p > c*sens))
#[num_y, num_x] = np.where((corrected_w < c*0.97) | (corrected_w > c*1.03))
ar = np.zeros((3,3))
ar2 = np.zeros((3,3))
#pool = multi.Pool(processes=6)
iter = num_y.shape[0]
for n in range(iter):
#parmap.map(bad_pixel_correction, [n, num_y, num_x, ar, ar2, corrected_w, corrected_org_I], pm_pbar=True, pm_processes=6)
for j in range(-1,2):
for k in range(-1,2):
if num_y[n]+j == -1 or num_x[n]+k == -1 or num_y[n]+j == 576 or num_x[n]+k == 576:
ar[j+1][k+1] = 0
ar2[j+1][k+1] = 0
else:
ar[j+1][k+1] = corrected_w[num_y[n]+j][num_x[n]+k]
ar2[j+1][k+1] = corrected_org_I[num_y[n]+j][num_x[n]+k]
ar[1][1] = 0
ar2[1][1] = 0
corrected_w[num_y[n]][num_x[n]] = np.sum(ar)/np.count_nonzero(ar)
corrected_org_I[num_y[n]][num_x[n]] = np.sum(ar2)/np.count_nonzero(ar2)
c = np.mean(corrected_w) # constant
## flat field correction
FFC = np.uint16(np.divide(c*(corrected_org_I-B), (corrected_w-B)))
F = np.fft.fft2(FFC)
Fshift = np.fft.fftshift(F)
magnitude_spectrum3 = 20*np.log(np.abs(Fshift))
[row, col] = org_I.shape
[row2, col2] = np.array([row, col], dtype=np.int) // 2
row2_range = 1
col2_range = 2
Fshift[:row2-row2_range-1, col2-col2_range-1:col2+col2_range] = 0
Fshift[row2+row2_range:, col2-col2_range-1:col2+col2_range] = 0
fishift = np.fft.ifftshift(Fshift)
result = np.fft.ifft2(fishift)
print("time :", time.time() - start)
cv2.imwrite(r'C:\Users\jhjoo\Desktop\corrected_org_I (' + str(i) + ').png', result)
cv2.imwrite(r'C:\Users\jhjoo\Desktop\corrected_org_I (' + str(i) + ').png', corrected_org_I)
cv2.imwrite(r'C:\Users\jhjoo\Desktop\corrected_w (' + str(i) + ').png', corrected_w)

cv2.imwrite takes first argument as a string, not a list. You should fix your code as following:
cv2.imwrite(r'C:\Users\Desktop\result (' + str(i) + ').png', result) #result is 16bit image

'numpy.ndarray' object is not callable in the Ipython console

I am trying to calculate cpmx, hmx, smpx, tmpx and smvx by doing simple interpolation after loading the data from excel into pandas dataframe.
While calling the function with cpmx=absmatdata(1,0,0,0,44.011,100) I see:
'numpy.ndarray' object is not callable
Any idea how to go about this?
Here is my code:
import numpy as np
import pandas as pd
def absmatdata(a,b,c,d,material,tmp_ref):
material_map = {2.016: 'H2', 28.016: 'N2', 32.000: 'O2', 32.065: 'S',
18.016: 'H2O', 64.065: 'SO2', 12.001: 'C Graphite',
28.011: 'CO', 44.011: 'CO2', 16.043: 'CH4', 30.070: 'C2H6',
44.097: 'C3H8', 58.124: 'C4H10'}
if material in material_map:
df = pd.read_excel('F:\MAschinenbau\Bachelorarbeit\ABSMAT.xlsx',sheet_name=material_map[material])
else:
print('No data for this material available')
df = [list(np.arange(0,1100,100)),list(np.arange(0,11,1)),list(np.arange(0,11,1)),list(np.arange(0,11,1)),list(np.arange(0,11,1))]
tmp = df.values[:,0]
cpm = df.values[:,1]
hm = df.values[:,2]
smp = df.values[:,3]
smv = df.values[:,4]
tn = np.size(df)
tmp0 = tmp_ref
tmpx = a
cpmx = 0
hmx = b
smpx = c
smvx = d
if a==0 and b==0 and c==0 and d==0:
print('All values are zero')
elif a!=0 and b==0 and c==0 and d==0:
print('T interpolation')
for i in range(0,tn-1):
if tmpx > tmp(i) and tmpx <= tmp(i+1):
int_fak = (tmpx-tmp(i))/(tmp(i+1)-tmp(i))
cpmx = cpm(i) + int_fak*(cpm(i+1)-cpm(i))
hmx = hm(i) + int_fak*(hm(i+1)-hm(i))
smpx = smp(i) + int_fak*(smp(i+1)-smp(i))
smvx = smv(i) + int_fak*(smv(i+1)-smv(i))
return tmpx, cpmx, hmx, smpx, smvx

You set df to DataFrame
You set tmp = df.values[:,0]
You have numpy.ndarry at tmp
You have to get its items with [] not with ()
Your loop part
if tmpx > tmp(i) and tmpx <= tmp(i+1):
int_fak = (tmpx-tmp(i))/(tmp(i+1)-tmp(i))
cpmx = cpm(i) + int_fak*(cpm(i+1)-cpm(i))
hmx = hm(i) + int_fak*(hm(i+1)-hm(i))
smpx = smp(i) + int_fak*(smp(i+1)-smp(i))
smvx = smv(i) + int_fak*(smv(i+1)-smv(i))
Should change with
if tmpx > tmp[i] and tmpx <= tmp[i+1]:
int_fak = (tmpx-tmp[i])/(tmp[i+1]-tmp[i])
cpmx = cpm[i] + int_fak*(cpm[i+1]-cpm[i])
hmx = hm[i] + int_fak*(hm[i+1]-hm(i))
smpx = smp[i] + int_fak*(smp[i+1]-smp[i])
smvx = smv[i] + int_fak*(smv[i+1]-smv[i])
Also you need to change your tn to
tn = np.size(df.values[:,0])

Python and Matlab code significantly slower then C++

I have to do a speed comparison of a simple program for detrended fluctuation analysis for Python, Matlab and C++. Unfortunately the speed of my Python and Matlab code was about two orders of magnitude slower then I expected. So either is what I read about those languages wrong, or I wrote the code in a very inefficient way.
That's why I'm asking if there are any obvious ways to speed up the code pasted below. You can download the Data here if you would like to run the code yourself.
Python hrv_dfa module
import numpy as np
import matplotlib.pyplot as plt
import math
import sys
import time
def hrv_dfa(file, version, verbose):
r_data = np.fromfile(file, dtype=float, count=-1, sep=" ")
# compute intervals
RR = RR_intervals_from_data(r_data)
RRavg = RR.mean(axis = 0)
y = (RR - RRavg).cumsum(axis = 0)
if (verbose):
fig = plt.ion().figure()
fig.ion()
fig.plot(y, 'o')
fig.show()
N = len(y)
#window size
lk = range(5, int(np.floor(N/10)+1))
E = np.zeros((len(lk), 1), dtype=float)
tic = time.time()
for k in range(0, len(lk)):
window = lk[k]
Sq_sum = 0
y_detrended = np.zeros((len(y), 1), dtype=float)
if (version == 1):
for i in range(0, N-window, window):
y_in_window = y[i:i+window]
x_in_window = range(i,i+window)
(a, b) = np.polyfit(x_in_window, y_in_window, 1)
y_regression = np.add(np.multiply(a, x_in_window), b)[np.newaxis]
y_detrended[i:i+window] = y_in_window - y_regression.T # bylo [i:i+window-1]
elif (version == 2):
for i in range(0, N-window):
#wybor danych z danego okna
x_in_window = range(i, i+window)
y_in_window = y[i:i+window]
#regresja liniowa dla danej probki
(a, b) = np.polyfit(x_in_window, y_in_window, 1)
#usuwanie trendu
if (not (window % 2 == 0)):
y_regression = a*(i+window/2+0.5)+b
y_detrended[i] = y[i+window/2+0.5] - y_regression
else:
y_regression1 = a*(i+window/2-0.5)+b
y_regression2 = a*(i+window/2+0.5)+b
y_detrended[i] = (y[i+window/2-0.5] - y_regression1 + y[i+window/2+0.5] - y_regression2)/2
else:
print( 'unsupported window version')
sys.exit([1])
if (verbose):
fig.plot(y_detrended, 'bl-')
fig.show()
Sq_sum = np.dot(y_detrended.T, y_detrended)
E[k] = np.sqrt(Sq_sum/N)
toc = time.time()
len(E)
if verbose:
fig = plt.figure()
x = np.log10(lk)
y = np.log10(E)
if (verbose):
plt.plot(x, y, 'o')
(a, b) = np.polyfit(x, y, 1)
if (verbose):
y_all_fitted = a*x+b
plt.plot(x, y_all_fitted)
#print lk[0:12]
(a1, b1) = np.polyfit(x[0:12], y[0:12], 1)
if (verbose):
y_short_term_fitted = a1*x[0:12]+b1
plt.plot(x[0:12], y_short_term_fitted)
(a2, b2) = np.polyfit(x[12:], y[12:], 1)
if (verbose):
y_long_term_fitted = a2*x[12:]+b2
plt.plot(x[12:], y_long_term_fitted)
if (verbose):
plt.show()
exec_time = toc - tic
print( "a = %f" % a)
print( "a1 = %f" % a1)
print( "a2 = %f" % a2)
print( "exec_time = %f" % exec_time)
#plt.show(block=True)
return a, a1, a2, exec_time
def RR_intervals_from_data(data):
RR = np.zeros((len(data), 1), dtype=float)
for t in range(len(data)-1):
RR[t] = (data[t+1]-data[t])
return RR
if __name__ == "__main__":
hrv_dfa(arg[1], arg[2], arg[3])
Python test code (runs every sample 10 times and saves results and execution time to csv)
import hrv_dfa34 as hrv_dfa
import sys
import csv
def test_hrv_dfa():
file_numbers = ["100", "101"] #, "102", "103", "104", "105", "106", "107", \
#"108", "109", "111", "112", "113", "114", "115", "116", \
#"117", "118", "119", "121", "122", "123", "124", "201", \
#"202", "203", "205", "207", "208", "209", "210", "212", \
#"213", "214", "215", "217", "219", "220", "221", "222", \
#"223", "228", "230", "231", "232", "233", "234"]
prefix = "..\Data\RPeakData"
file_names = []
for i in range(0, len(file_numbers)):
file_names.append(prefix + file_numbers[i] + ".txt")
for file_number_idx in range(0, len(file_numbers)):
for test_number in range(0, 10):
a, a1, a2, exec_time = hrv_dfa.hrv_dfa(file_names[file_number_idx], 1, False)
with open('fwrite_sw.csv', 'a', newline='') as fwrite_sw:
writer = csv.writer(fwrite_sw, delimiter=';', \
quotechar='"', quoting=csv.QUOTE_MINIMAL)
writer.writerow([file_numbers[file_number_idx], a[0], a1[0], a2[0], exec_time])
#fwrite_sw.write( + " " + a + " " + a1 + " " + a2 + " " + exec_time + "\n")
for file_number_idx in range(0, len(file_numbers)):
for test_number in range(0, 10):
a, a1, a2, exec_time = hrv_dfa.hrv_dfa(file_names[file_number_idx], 2, False)
with open('fwrite_mw.csv', 'a', newline='') as fwrite_mw:
writer2 = csv.writer(fwrite_mw, delimiter=';', \
quotechar='"', quoting=csv.QUOTE_MINIMAL)
writer2.writerow([file_numbers[file_number_idx], a[0], a1[0], a2[0], exec_time])
#fwrite_mw.write(file_numbers[file_number_idx] + " " + a + " " + a1 + " " + a2 + " " + exec_time + "\n")
print('\nDone!')
if __name__ == '__main__':
test_hrv_dfa()
Matlab hrv_dfa module
function [ alpha, alpha1, alpha2, exec_time ] = hrv_dfa(file, version, verbose)
% skrypt ładujący dane i uruchamiający poszczególne metody hrv_dfa
% metoda 1 - stałe okno (obcinana nadmiarowa część próbek)
% metoda 2 - ślizgające się okno
fileID = fopen(file, 'r');
formatSpec = '%f';
r_data = fscanf(fileID, formatSpec);%cumsum(r_intervals) % << sztuczne dane
fclose(fileID);
%save r_data.out r_data -ASCII
RR = RR_intervals_from_data(r_data);
RRavg = mean(RR);
y = cumsum(RR-RRavg);
N = length(y);
lk = 5:floor(N/10);
E = zeros(1,length(lk));
tic;
for k = 1:length(lk)
window = lk(k);
Sq_sum = 0;
y_detrended = zeros(size(y));
if version == 1
for i = 1:window:(N-window+1)
y_in_window = y(i:i+window-1);
x_in_window = (i:i+window-1)';
linear_regression_coeffs = polyfit(x_in_window, y_in_window,1);
a = linear_regression_coeffs(1);
b = linear_regression_coeffs(2);
y_regression = a*x_in_window+b;
y_detrended(i:i+window-1) = y_in_window - y_regression;
end
elseif version == 2
for i = 1:N-window
x_in_window = (i:i+window-1)';
y_in_window = y(x_in_window);
linear_regression_coeffs = polyfit(x_in_window, y_in_window,1);
a = linear_regression_coeffs(1);
b = linear_regression_coeffs(2);
if (~(mod(window, 2)==0))
y_regression = a*(i+window/2 - 0.5)+b;
y_detrended(i) = y(i + window/2 - 0.5) - y_regression;% +window/2
else
y_regression1 = a*(i+window/2-1)+b;
y_regression2 = a*(i+window/2)+b;
y_detrended(i) = (y(i+window/2-1) - y_regression1 + y(i+window/2) - y_regression2)/2;
end
end
else
exit('unsupported window version');
end
if verbose == true
figure(99);
plot(y)
hold on;
plot(y_detrended)
end
Sq_sum = y_detrended'*y_detrended;
E(k) = sqrt(Sq_sum/N);
end
exec_time = toc;
x = log10(lk);
y = log10(E);
if verbose == true
figure(999);
hold off
plot(x, y, '.')
end
allRegression = polyfit(x,y,1);
alpha = allRegression(1);
beta = allRegression(2);
y_all_fitted = alpha*x+beta;
shortTermRegression = polyfit(x(1:12), y(1:12), 1);
alpha1 = shortTermRegression(1);
beta1 = shortTermRegression(2);
y_short_term_fitted = alpha1*x(1:12)+beta1;
longTermRegression = polyfit(x(12:end), y(12:end), 1);
alpha2 = longTermRegression(1);
beta2 = longTermRegression(2);
y_long_term_fitted = alpha2*x(12:end)+beta2;
if verbose == true
hold on;
h1 = plot(x,y_all_fitted, 'k-', 'LineWidth', 2.5);
h2 = plot(x(1:12), y_short_term_fitted, 'g-', 'LineWidth', 2.5);
h3 = plot(x(12:end), y_long_term_fitted, 'r-', 'LineWidth', 2.5);
legend([h1 h2 h3], {'prosta aproksymująca dla wszystkich lk', ...
'prosta aproksymująca dla log10(5) <= lk <=log10(16)', ...
'prosta paroksymująca dla log10(16) < lk'}, 'Location', 'northwest');
xlabel('log10(lk)');
ylabel('log10(E(lk))');
grid on;
hold off;
end
% helper - returns intervals
function [RR] = RR_intervals_from_data(data)
RR = zeros(size(data));
for t = 1:(size(data)-1)
RR(t) = data(t+1)-data(t);
end
end
end
Matlab test code (runs every sample 10 times and saves to csv)
file_numbers = ['100'; '101'; '102'; '103'; '104'; '105'; '106'; '107'; ...
'108'; '109'; '111'; '112'; '113'; '114'; '115'; '116'; ...
'117'; '118'; '119'; '121'; '122'; '123'; '124'; '201'; ...
'202'; '203'; '205'; '207'; '208'; '209'; '210'; '212'; ...
'213'; '214'; '215'; '217'; '219'; '220'; '221'; '222'; ...
'223'; '228'; '230'; '231'; '232'; '233'; '234'];
prefix = '..\Data\RPeakData';
file_names = [];
for i = 1:size(file_numbers, 1)
file_names = [file_names; [prefix, file_numbers(i, :), '.txt']];
end
for file_number_idx = 1:size(file_numbers, 1)
for test_number = 1:10
[a, a1, a2, exec_time] = hrv_dfa(file_names(file_number_idx, :), 1, false);
toStore = [str2num(file_numbers(file_number_idx,:)), a, a1, a2, exec_time];
dlmwrite('fmwrite_sw.csv', toStore, '-append', 'delimiter', ';');
end
end
for file_number_idx = 1:length(file_numbers)
for test_number = 1:10
[a, a1, a2, exec_time] = hrv_dfa(file_names(file_number_idx, :), 2, false);
toStore = [str2num(file_numbers(file_number_idx,:)), a, a1, a2, exec_time];
dlmwrite('fmwrite_mw.csv', toStore, '-append', 'delimiter', ';');
end
end
display('Done!');