def PlotPersist(dtsetfile, plotdir, grpname):
dtset = np.load(dtsetfile, allow_pickle=True)
#allow_pickle op is for adapting spec change of numpy 1.16.3 and later
dts = dtset['dataset']
dataset0 = []
for dt in dts:
dt0 = dt['inputs/0']
dataset0.append(dt0)
dim0 = len(dataset0)
dim1 = len(dataset0[0])
dim2 = len(dataset0[0][0])
PCdata = []
for dt64 in dataset0:
for dt in dt64:
PCdata.append([dt[0], dt[1], dt[2]])
#Plot persistent diagram
hc.PDList.from_alpha_filtration(PCdata,
save_to="pointcloud.idiagram",
save_boundary_map=True)
pdlist = hc.PDList("pointcloud.idiagram")
pd = pdlist.dth_diagram(1)
pd.histogram().plot(colorbar={"type": "log"})
plotfile1 = plotdir + grpname + "-pers.png"
plt.savefig(plotfile1)
plt.close()
print(
f'Plot PC1/PC2/PC3 data of {grpname}: Shape= {dim0} x {dim1} x {dim2}')
def make_PD(pointcloud):
hc.PDList.from_alpha_filtration(pointcloud,
save_to="pointcloud.idiagram",
save_boundary_map=True)
pdlist = hc.PDList("pointcloud.idiagram")
pd = [pdlist.dth_diagram(i) for i in range(3)]
for j in range(len(pd)):
lengths = [
math.sqrt(pd[j].deaths[i]**2 + pd[j].births[i]**2)
for i in range(len(pd[j].deaths))
]
length_maximum = max(lengths)
for i in range(len(pd[j].deaths)):
pd[j].births[i] = math.sqrt(2) * pd[j].births[i] / length_maximum
pd[j].deaths[i] = math.sqrt(2) * pd[j].deaths[i] / length_maximum
return pd
def make_PD(pointcloud):
hc.PDList.from_alpha_filtration(pointcloud,
save_to="pointcloud.idiagram",
save_boundary_map=True)
pdlist = hc.PDList("pointcloud.idiagram")
PD = []
for j in range(3):
pd = pdlist.dth_diagram(j)
death_max = max(pd.deaths)
birth_max = max(pd.births)
#0-1区間で正規化
for i in range(len(pd.deaths)):
pd.births[i] = pd.births[i] / birth_max if pd.births[i] != 0 else 0
pd.deaths[i] = pd.deaths[i] / death_max if pd.deaths[i] != 0 else 0
PD.append(pd)
return PD
#以下実行部
data = data_processing(path)
death_birth = []
for i in range(len(data_ranges)):
# for channel,pic_name,txt_name in zip(channels,picture_names,output_text_name):
data = data_processing(path)
pointcloud = delay(data, 'channel2', data_ranges[i])
#PD生成・保存
hc.PDList.from_alpha_filtration(pointcloud,
save_to="pointcloud.idiagram",
save_boundary_map=True)
pdlist = hc.PDList("pointcloud.idiagram")
pd = [pdlist.dth_diagram(i) for i in range(3)] #0~2次元のPDを生成
pd = normalization(pd) #データの正規化
betti_sequence = generating_betti_sequence(pd)
# output_betti_picture(betti_sequence)
output_betti_text(betti_sequence, "test{0}.txt".format(i + 1))
# death_birth = calc_death_birth(pd,death_birth,i)
# output_picture(pd,picture_names[i])
# output_text(pd)
print("Done")
#compute R-wave peak coordinates
rpeaks = signal.find_peaks(ekg[:, 1], height=0.5, distance=10)
r_peak_xcs, r_peak_xc_idx = get_rpeak_xcs(rpeaks, ekg)
rr_int_avg, rr_int_sd = get_rr_interval(r_peak_xcs)
#trim signal to begin and end with an R-wave peak
ekg = trim(ekg, r_peak_xcs)
#compute persistent homology of processed signal
output = hc.PDList.from_alpha_filtration(ekg,
no_squared=True,
save_boundary_map=True,
save_phtrees=True,
save_to="pointcloud.pdgm")
pd1 = hc.PDList("pointcloud.pdgm").dth_diagram(1)
persist = np.asarray(pd1.deaths - pd1.births)
births = np.asarray(pd1.births)
deaths = np.asarray(pd1.deaths)
#compute optimal cycle representatives
#cycle_xcs,cycle_ycs,bps=get_vol_opt_cycle_centroid_coords(persist,pd1)
cycle_xcs, cycle_ycs, bps = get_card_opt_cycle_centroid_coords(
persist, pd1)
#cycle_xcs,cycle_ycs,bps=get_stab_vol_cycle_centroid_coords(persist,pd1)
#cycle_xcs,cycle_ycs,bps=get_stab_subvol_cycle_centroid_coords(persist,pd1)
#measure intervals of interest and compute the index locations of P,Q,S, and T-waves within the 1d array persist
int_tda[i, :], idx_p, idx_q, idx_s, idx_t = get_intervals_and_H1wave_idxs(
ekg, r_peak_xcs, rr_int_avg, persist, births, cycle_xcs, cycle_ycs,
bps)