def get_data():
global RES, INPUT_RES, OUTPUT_RES
m_x = []
m_y = []
for object_index in range(120):
for series_index in range(24):
series = []
for i in range(3):
path = f"coil-20/obj{object_index // 6 + 1}__{series_index * 3 + i}.png" if object_index % 6 == 0 else \
f"coil-100/obj{object_index - (object_index + 1) // 6 + 1}__{(series_index * 3 + i) * 5}.png"
image = resample(cv.imread(path, cv.IMREAD_GRAYSCALE), INPUT_RES / RES, 0).astype('float32') / 255
series.append(image)
if i == 1:
image = resample(cv.imread(path, cv.IMREAD_GRAYSCALE), OUTPUT_RES / RES, 0).astype('float32') / 255
m_y.append(image)
m_x.append(series)
m_x_train, m_x_test, m_y_train, m_y_test = train_test_split(m_x, m_y, test_size=0.15, shuffle=False)
m_x_train = np.array(m_x_train)
m_y_train = np.array(m_y_train)
m_x_test = np.array(m_x_test)
m_y_test = np.array(m_y_test)
m_y_train = m_y_train.reshape((m_y_train.shape[0], OUTPUT_RES * OUTPUT_RES))
m_y_test = m_y_test.reshape((m_y_test.shape[0], OUTPUT_RES * OUTPUT_RES))
m_x = np.array(m_x)
m_y = np.array(m_y)
m_y = m_y.reshape((m_y.shape[0], OUTPUT_RES * OUTPUT_RES))
return m_x, m_y, m_x_train, m_x_test, m_y_train, m_y_test
feat_df = feat_df[~feat_df.index.duplicated(keep='last')]
print("Pre-Processing Runtime: " + str(time.time() - start_time)) # Loop Runtime
# =============================================================================
# RESAMPLING
# =============================================================================
target_hz = 40
overlap = 2
window_size = target_hz * 2
hop = int(window_size/overlap)
sep = 'Sensus OS' # 'Device ID' or 'Sensus OS'
datetime_range = [['2018-9-13 17:31:35','2018-9-13 17:48:35'],['2018-9-14 14:34:32','2018-9-14 15:49:19'],['2018-9-14 19:27:19','2018-9-14 19:57:48']] # enter precise data collection datetime ranges
start_time = time.time()
resamp_df, keys = re.resample(feat_df,target_hz,sep,datetime_range)
print("Resampling Runtime: " + str(time.time() - start_time)) # Loop Runtime
# =============================================================================
# FEATURE EXTRACTION
# =============================================================================
target_path = '/Users/sm7gc/Desktop/WASH/Featurized/09-14-18/OS&Participant' if sep == 'Sensus OS' else '/Users/sm7gc/Desktop/WASH/Featurized/09-14-18/Device'
shutil.rmtree(target_path)
os.makedirs(target_path)
features = {} # dictionary for calculated features
for k in range(len(keys)):
if __name__ == "__main__":
import read_segy as rsegy
import resampling as rs
segy = _read_segy('data/seis/CX_ZJ_ori.SGY', headonly=True)
# time_start = 900.0
# time_stop = 1500.0
# time_step = 2.0 # ms
# inline = 3662
# xline = 1938
trace = rsegy.extract_seismic(segy, line=3662, xline=1938,
time_start = 810, time_stop= 1500, time_step=2, name='amp')
trace_res = rs.resample(trace.time, trace.amp,
time_start_re = 820,
time_stop_re = 1400,
time_step_re = 1,
kind="quadratic", plot=False,
logname='amp')
cc,ff = spectrum_cwt(trace_res.time, trace_res.amp/10000,
colorscale=1, widths=100, wavelet='morl', freqmax=100)
freqs,amplitude = spectrum_fft_filter(trace_res.time, trace_res.amp/10000,
fft_size = 2000,
freq_set=True, freq_min=0, freq_max=100,
withfilter=False, filter_value=60, filter_slop=10,
btype='low', analog=False, output='ba')
plt.figure(figsize=(3,8))
plt.plot(trace_res.amp/10000, trace_res.time, 'k') #'steelblue'
plt.xlabel(u"Amplitude")
plt.ylabel(u"Time(ms)")
import faux
import resampling
import os
import numpy as np
# Pacientes id
id = 1
data_path = "C:/Users/Darwin/Documents/NSCLC-RADIOMICS-INTEROBSERVER1/interobs06/02-18-2019-CT-43086/01133/"
# Carpeta donde se almacenara los resultados
output_path = "C:/Users/Darwin/Documents/ResultadosNSCL/"
patient = faux.load_scan(data_path)
imgs_to_process = np.load(output_path + 'fullimages_{}.npy'.format(id))
print("Slice Thickness: %f" % patient[0].SliceThickness)
print("Pixel Spacing (row, col): (%f, %f) " %
(patient[0].PixelSpacing[0], patient[0].PixelSpacing[1]))
print("Shape before resampling\t", imgs_to_process.shape)
imgs_after_resamp, spacing = resampling.resample(imgs_to_process, patient,
[1, 1, 1])
print("Shape after resampling\t", imgs_after_resamp.shape)
np.save(output_path + "imgs_after_resamp_%d.npy" % (id), imgs_after_resamp)
def IDBMIOT(S, A, k1, k2):
#c, p1, p2, m, D <- S & A
c = pd.Series.sort_values(S['Class'].value_counts())
p1 = len(A.numeric)
p2 = len(A.nominal)
m = len(c)
D = c
#resampling, get orate(c)
orates = resample(c, D, S)
orate = orates.orate
print("Oversampling rate: " + str(orate))
S = orates.samples
S = S.dropna()
c = pd.Series.sort_values(S['Class'].value_counts())
m = len(c)
D = c
minorityClasses = c.drop(c.index[m - 1])
c = minorityClasses
retransform_sample = {}
#iterate on class
for c1 in c.keys():
#instance of class c1
xs = S[S['Class'] == c1]
class_sample_instance = len(xs)
samples = chooseSample(S, c1, k1, k2)
_sample = samples
"""#_samples computation
#_sample = pd.get_dummies(samples, columns=["Class"], prefix=["class"])
#mean
#us = xs.mean(axis=0)
#eclipse curve
ec = 0
it = 0
for i in class_sample_instance:
ec = ec + math.pow((xs.iloc(it) - us),2)
it = it + 1
os = math.sqrt( (1/class_sample_instance) * ec)
#normalization
_sample = (_sample - us)/os
#compute N & M
N = 1/class_sample_instance
adiag = []
g = 0
while g < len(A.numeric):
adiag.append(1)
g = g + 1
h = 0
while h < (m * len(A.nominal)):
k=0
while k<len(c.keys()):
v = _sample.iloc[len(A.numeric)+k]
adiag.append(2/len(v))
k = k + 1
h = h +1
M = np.diag(adiag) """
#get features and label
X = _sample.drop('Class', axis=1)
y = _sample['Class']
#Standardise
x_std = StandardScaler().fit_transform(X)
#coefficent of vector varriance
features = x_std.T
covariance_matrix = np.cov(features)
V = covariance_matrix
#convert samples into Principle component
pca = PCA(n_components=2)
principalComponents = pca.fit_transform(x_std)
principalDf = pd.DataFrame(
data=principalComponents,
columns=['principal component 1', 'principal component 2'])
#principalDf['Class'] = _sample['Class']
cf = _sample[['Class']]
#finalDf = pd.append([principalDf, _sample[['Class']]], axis = 1)
finalDf = principalDf.join(cf)
newSampleInPC = pd.DataFrame()
newSampleInOC = pd.DataFrame()
#sample generation
if (orate[c1] > 0):
j = 0
while j < orate[c1]:
#random sample choose
maxValue = finalDf[finalDf['Class'] == c1].max()[0]
randomSample = finalDf[np.isclose(
finalDf['principal component 1'], maxValue)]
#cal random sample
rs = sum(random.sample(range(1, p1 + m + k2), p1 + m))
randomSample['principal component 1'] = randomSample[
'principal component 1'] + rs
newSampleInPC = newSampleInPC.append(randomSample)
j = j + 1
if not newSampleInPC.empty:
ff = newSampleInPC.drop('Class', axis=1)
# get new sample
newSampleInOC = pca.inverse_transform(ff)
# transform
column_label = X.columns
retransform_sample = pd.DataFrame(newSampleInOC,
columns=column_label)
retransform_sample['Class'] = c1
#Addition of resamples into orginal sample
""" if retransform_sample.empty is False:
#print("No new sample were generated!")
raise Exception('No new sample were generated!')
else: """
S.append(retransform_sample, ignore_index=True)
#return samples
return S