def draw_rectangle(image_in,width,height,xlb,xub,ylb,yub,image_out,color):
# Drawing a rectangle in a picture
dpi = 80
im = Image.open(image_in)
ratiox = im.size[0]/width
ratioy = im.size[1]/height
xlb = xlb*ratiox
xub = xub*ratiox
ylb = ylb*ratioy
yub = yub*ratioy
# What size does the figure need to be in inches to fit the image?
figsize = width / float(dpi), height / float(dpi)
figure = plt.figure(figsize=figsize)
ax = figure.add_axes([0, 0, 1, 1])
# Hide spines, ticks, etc.
ax.axis('off')
rect = matplotlib.patches.Rectangle((xlb,im.size[1]-yub),xub-xlb,yub-ylb, edgecolor=color, facecolor="none")
ax.imshow(im)
ax.add_patch(rect)
figure.savefig(image_out,dpi=dpi,transparent=True)
return 0
def PCA_expansion(iq_mat, config):
'''
Calculate the PCA of the vector and generate samples from it
'''
mean = np.mean(iq_mat, axis=0)
standardized_iq_mat = iq_mat - np.mean(iq_mat, axis=0)
cov_mat = np.cov(standardized_iq_mat) # Calculate the covariance matrix
cov_mat = np.zeros((iq_mat.shape[0], iq_mat.shape[0]))
for index in range(iq_mat.shape[1]):
cov_mat += np.matmul(
iq_mat[:, index] - mean,
np.transpose(iq_mat[:, index] - mean)) / iq_mat.shape[1]
'''
Eigen Value decomposition
'''
eigen_values, eigen_vector = np.linalg.eig(
cov_mat) # Calculate the eigen values and eigen vectors
idx = np.argsort(eigen_values)[::-1]
eigen_vector = eigen_vector[:, idx]
'''
Generate new samples
'''
NUMBER_OF_KL_COEEFIECENT = config.number_of_pca_coeff
VARIANCE_SCALING = config.pca_augmentation_scaling
KL_mat = np.matmul(np.transpose(standardized_iq_mat), eigen_vector)
KL_mat[:, :
NUMBER_OF_KL_COEEFIECENT] = KL_mat[:, :
NUMBER_OF_KL_COEEFIECENT] + VARIANCE_SCALING * np.multiply(
KL_mat[:, :
NUMBER_OF_KL_COEEFIECENT],
np.random.randn(
32,
NUMBER_OF_KL_COEEFIECENT))
new_iq_mat = np.matmul(eigen_vector, np.transpose(KL_mat))
figure, axes = plt.subplots(nrows=2, ncols=2)
axes[0, 0].imshow(standardized_iq_mat)
axes[0, 0].set_title('Freq Response Org')
axes[0, 1].imshow(np.real(new_iq_mat))
axes[0, 1].set_title('Freq Response Generated 1 ')
axes[1, 0].imshow(np.real(new_iq_mat))
axes[1, 0].set_title('Freq Response Generated 2 ')
axes[1, 1].imshow(np.real(new_iq_mat))
axes[1, 1].set_title('Freq Response Generated 3 ')
figure.savefig('test_pca')
return new_iq_mat
def Sample_rectangle_from_spectrogram(iq_mat, config):
NUM_OF_ROWS = config.rect_augment_num_of_rows
NUM_OF_COLS = config.rect_augment_num_of_cols
GAP_OF_DOPPLER_BURST_FROM_EDGE = config.rect_augment_gap_doppler_burst_from_edge
MAX_NUM_OF_BINS = 126
MAX_NUM_OF_TIMESTEPS = 32
new_iq_list = []
doppler_burst = np.argmax(
iq_mat, axis=0) # Extract the doppler burst from the matrix
for row_ind in range(MAX_NUM_OF_BINS - NUM_OF_ROWS):
for col_ind in range(MAX_NUM_OF_TIMESTEPS - NUM_OF_COLS):
for col_sub_ind in range(col_ind, col_ind + NUM_OF_COLS):
'''
Check that we have Doppler Burst with a gap of GAP_OF_DOPPLER_BURST_FROM_EDGE from the edge from all options
( if not discard this rect option )
'''
discard_rect_flag = False
if doppler_burst[col_sub_ind] > (row_ind + NUM_OF_ROWS - GAP_OF_DOPPLER_BURST_FROM_EDGE)\
or doppler_burst[col_sub_ind] < (row_ind + GAP_OF_DOPPLER_BURST_FROM_EDGE):
discard_rect_flag = True
break
if discard_rect_flag:
continue
new_iq_list.append(iq_mat[row_ind:row_ind + NUM_OF_ROWS,
col_ind:col_ind + NUM_OF_COLS])
figure, axes = plt.subplots(nrows=2, ncols=2)
axes[0, 0].imshow(iq_mat)
axes[0, 0].set_title('Freq Response Org')
index = np.random.choice(range(len(new_iq_list)))
axes[0, 1].imshow(new_iq_list[index])
axes[0, 1].set_title('Freq Response Extracted index'.format(index))
index = np.random.choice(range(len(new_iq_list)))
axes[1, 0].imshow(new_iq_list[index])
axes[1, 0].set_title('Freq Response Extracted {}'.format(index))
index = np.random.choice(range(len(new_iq_list)))
axes[1, 1].imshow(new_iq_list[index])
axes[1, 1].set_title('Freq Response Extracted {} '.format(index))
figure.savefig('test_rect_sample')
return new_iq_list
def Calculate(self, event):
#Get Occupation selections and populate a list
item_occ = self.tree_occ.GetSelections()
self.calc_occ_list = listSetup(self, item_occ, self.tree_occ)
#Get Territory selections and populate a list
item_terr = self.tree_terr.GetSelections()
self.calc_terr_list = listSetup(self, item_terr, self.tree_terr)
if self.calc_occ_list != [] and self.calc_terr_list != []:
#Generate the lists to be used for charts
chart, other_chart = listGenerateCharts(self, self.calc_occ_list, self.calc_terr_list)
#Sort the lists generated
sorted_chart = chart
sorted_other_chart = other_chart
#Generate the charts
figure = GenerateFigure(self, sorted_chart)
figure.savefig('current.png')
if sorted_other_chart != []:
figure_other = GenerateFigure(self, sorted_other_chart)
figure_other.savefig('current_other.png')
grp.plot.bar()
grp.to_csv("CountryVsIncome.csv")
####Countplot
xx = train_df[['NativeCountry', 'Income']]
xx = xx[xx.NativeCountry != -999]
figure()
countplot(data=xx, y=xx.Income)
show()
xx = train_df[['Gender', 'Income']]
xx = xx[xx.Gender != -999]
#figure(figsize=(12,6))
svm = countplot(data=xx, x=xx.Income, hue="Gender")
figure = svm.get_figure()
figure.savefig('gender_income.png', dpi=400)
show()
xx = train_df[['WorkClass', 'Income']]
xx = xx[xx.WorkClass != -999]
svm = countplot(data=xx, x=xx.Income, hue="WorkClass")
figure = svm.get_figure()
figure.savefig('Work_income.png', dpi=400)
show()
xx = train_df[['HoursPerWeek', 'Income', 'NativeCountry']]
xx = xx[xx.NativeCountry != -999]
#yy=xx[xx['WorkClass']=='Private']
