def main():
ground_truth_path = './data/cran_Ground_Truth.tsv'
search_engine_conf_path = './data/total_query_results.json'
configuration_path = './data/SearchEngines.csv'
ground_truth_dict = utils.read_ground_truth(ground_truth_path)
search_engine_conf = utils.read_json(search_engine_conf_path)
MRR_results = MRR(ground_truth_dict, search_engine_conf)
R_precision_results = R_Precision(ground_truth_dict, search_engine_conf)
configurations = pd.read_csv(configuration_path)
configurations['MRR'] = MRR_results
configurations['Mean'] = np.mean(R_precision_results, axis=1)
configurations['Min'] = np.min(R_precision_results, axis=1)
configurations['Max'] = np.max(R_precision_results, axis=1)
configurations['Median'] = np.median(R_precision_results, axis=1)
configurations['1_quartile'] = np.quantile(a=R_precision_results,
q=.25,
axis=1)
configurations['3_quartile'] = np.quantile(a=R_precision_results,
q=.75,
axis=1)
configurations_top_5 = configurations.sort_values(
by=['MRR'], ascending=False).head(5).SE_ID
print(list(configurations_top_5))
col_names = ['Conf_' + str(i) for i in list(configurations_top_5)]
#col_names.reverse()
search_engine_conf_top_5 = {
key: search_engine_conf[key]
for key in configurations_top_5
}
print('P@k....')
P_at_k_res = P_at_k(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(P_at_k_res).transpose()
temp_df.columns = col_names
ax = temp_df.plot(title='P@k')
ax.set_xlabel('k values')
ax.set_ylabel('Mean P@k')
plt.savefig('./Report/Images/Pk.png')
print('NCDG@k....')
ncdg_at_k_res = ncdg(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(ncdg_at_k_res).transpose()
temp_df.columns = col_names
ax = temp_df.plot(title='NCDG@k')
ax.set_xlabel('k values')
ax.set_ylabel('Mean NCDG@k')
plt.savefig('./Report/Images/NCDGk.png')
configurations = configurations.sort_values(by=['MRR'], ascending=False)
configurations.to_csv(r'./data/SearchEnginesResults.csv', index=False)
def rename_sudokus():
with open('ground_truth.renamed.csv', 'w', encoding='utf8') as f:
for i, (file_path,
coords) in enumerate(read_ground_truth(GT_OUT_FILE)):
i += 31
new_path = os.path.join(os.path.dirname(file_path),
f'sudoku_{i:d}.jpg')
if os.path.exists(new_path):
raise RuntimeError()
os.rename(file_path, new_path)
cells = [new_path] + [str(a) for a in coords.flatten()]
line = ', '.join(cells) + '\n'
print(line, end='')
f.write(line)
def community_search_for_all_nodes(graph,
ground_truth_file_address,
start_with_given_node=True):
"""do community search for all nodes separately, calculate accuracy measures then report the Avg. and SD.
Args:
graph ([nx.Graph]): [the given network]
ground_truth_file_address ([str]): [filename of the ground-truth information of communities]
start_with_given_node (bool, optional): [if Ture: start expansion with the given node, if False: start with a node of highest degree]. Defaults to False.
"""
start_time = time.time()
performance_info = dict()
ground_truth_com2nodes = utils.read_ground_truth(ground_truth_file_address)
for e, node in enumerate(graph.nodes()):
performance_info[node] = {
'degree': graph.degree[node],
'precision': 0.0,
'recall': 0.0,
'f1-score': 0.0
}
community = community_search(graph, node, start_with_given_node)
utils.update_performance_info(node, performance_info, community,
ground_truth_com2nodes)
precision = sum(performance_info[x]['precision']
for x in list(graph.nodes())) / graph.number_of_nodes()
recall = sum(performance_info[x]['recall']
for x in list(graph.nodes())) / graph.number_of_nodes()
f1_score = sum(performance_info[x]['f1_score']
for x in list(graph.nodes())) / graph.number_of_nodes()
sd = sqrt(
sum((performance_info[x]['f1_score'] - f1_score)**2
for x in list(graph.nodes())) / graph.number_of_nodes())
print('precision = %.4f' % precision, end='\t')
print('recall = %.4f' % recall, end='\t')
print('f1-score = %.4f' % f1_score, end='\t')
print('sd(fscore) = %.4f' % sd, end='\t')
finish_time = time.time()
print('time = %.4f' % (finish_time - start_time))
import cv2 as cv
import numpy as np
import config
from sudoku_detector import SudokuDetector
from utils import rotation_correction, read_ground_truth, show
RED = (0, 0, 255)
GREEN = (0, 255, 0)
BLUE = (255, 0, 0)
CYAN = (255, 255, 0)
MAGENTA = (255, 0, 255)
gt_annoatations = read_ground_truth(config.sudokus_gt_path)
detector = SudokuDetector()
for sudoku_index, (file_path, gt_coords) in enumerate(gt_annoatations):
if sudoku_index < 0:
continue
start = time.time()
sudoku_img_org = cv.imread(file_path)
# Ensure that the sudoku is always rotated by at most 45 deg in either direction.
sudoku_img_org, gt_coords = rotation_correction(sudoku_img_org, gt_coords)
det = detector.detect(sudoku_img_org)
def main():
ground_truth_path = './data/part_1_2__Ground_Truth.tsv'
result_se1_path = './data/part_1_2__Results_SE_1.tsv'
result_se2_path = './data/part_1_2__Results_SE_2.tsv'
result_se3_path = './data/part_1_2__Results_SE_3.tsv'
ground_truth = utils.read_ground_truth(ground_truth_path)
se1 = utils.read_result_se(result_se1_path)
se2 = utils.read_result_se(result_se2_path)
se3 = utils.read_result_se(result_se3_path)
search_engine_conf = {1: se1, 2: se2, 3: se3}
print('P@k....')
P_at_k_res = EvaluationMetrics.P_at_k(ground_truth,
search_engine_conf,
k_vals=[4])
PK = []
for k, P_k_list in P_at_k_res.items():
count = 1
for pk in P_k_list:
SE = 'SE' + str(count)
PK.append([SE, pk])
count += 1
df = pd.DataFrame(PK, columns=['SE', 'PK'])
print(P_at_k_res)
plt.figure(figsize=(15, 7))
plot = sns.barplot(x='SE', y='PK', data=df)
for p in plot.patches:
plot.annotate(format(p.get_height(), ',.2f'),
(p.get_x() + p.get_width() / 2., p.get_height()),
ha='center',
va='center',
xytext=(0, 10),
textcoords='offset points')
plt.title('P@k Evaluation Metrics of Search Engines with k =4')
plt.xlabel('Search Engines')
plt.ylabel('The average P@k over all provided queries')
plt.ylim(0, df['PK'].max() * 1.2)
plt.savefig('./Report/Images/Pk_part1_2.png')
print('R@k....')
R_at_k_res = EvaluationMetrics.R_at_k(ground_truth,
search_engine_conf,
k_vals=[4])
RK = []
for k, R_k_list in R_at_k_res.items():
count = 1
for rk in R_k_list:
SE = 'SE' + str(count)
RK.append([SE, rk])
count += 1
print(R_at_k_res)
df = pd.DataFrame(RK, columns=['SE', 'RK'])
print(df)
plt.figure(figsize=(15, 7))
plot = sns.barplot(x='SE', y='RK', data=df)
for p in plot.patches:
plot.annotate(format(p.get_height(), ',.2f'),
(p.get_x() + p.get_width() / 2., p.get_height()),
ha='center',
va='center',
xytext=(0, 10),
textcoords='offset points')
plt.title('R@k Evaluation Metrics of Search Engines with k =4')
plt.xlabel('Search Engines')
plt.ylabel('The average R@k over all provided queries')
plt.ylim(0, df['RK'].max() * 1.2)
plt.savefig('./Report/Images/Rk_part1_2.png')
print('NCDG@k....')
ncdg_at_k_res = EvaluationMetrics.ncdg(ground_truth,
search_engine_conf,
k_vals=[4])
nDCG_results = []
for k, nDCG_list in ncdg_at_k_res.items():
count = 1
for nDCG in nDCG_list:
SE = 'SE' + str(count)
nDCG_results.append([SE, nDCG])
count += 1
df = pd.DataFrame(nDCG_results, columns=['SE', 'nDCG'])
print(df)
plt.figure(figsize=(15, 7))
plot = sns.barplot(x='SE', y='nDCG', data=df)
for p in plot.patches:
plot.annotate(format(p.get_height(), ',.2f'),
(p.get_x() + p.get_width() / 2., p.get_height()),
ha='center',
va='center',
xytext=(0, 10),
textcoords='offset points')
plt.title('nDCG Evaluation Metrics of Search Engines with k =4')
plt.xlabel('Search Engines')
plt.ylabel('The average nDCG over all provided queries')
plt.ylim(0, df['nDCG'].max() * 1.2)
plt.savefig('./Report/Images/nDCG_part1_2.png')
import cv2 as cv
import config
from utils import rotation_correction, read_ground_truth
if __name__ == '__main__':
for file_name, coords in read_ground_truth(config.sudokus_gt_path):
img = cv.imread(file_name, cv.IMREAD_COLOR)
img, coords = rotation_correction(img, coords)
print(f'{file_name}, Shape: {img.shape}')
img = cv.polylines(img, [coords], True, (0, 255, 0), thickness=5)
cv.line(img,
tuple(coords[0, :]),
tuple(coords[1, :]), (255, 0, 255),
thickness=10)
cv.line(img,
tuple(coords[1, :]),
tuple(coords[2, :]), (255, 255, 0),
thickness=10)
h, w = img.shape[:2]
cx = int(round(w / 2))
cy = int(round(h / 2))
img = cv.drawMarker(img, (cx, cy), (0, 0, 255), thickness=5)
img = cv.resize(img, (1024, int(img.shape[0] / (img.shape[1] / 1024))))
def main():
# CRANFIELD
positions = (1, 3, 5, 10)
ground_truth_path = '../../../data/cran_Ground_Truth.tsv'
search_engine_conf_path = '../../../data/total_query_results_cran.json'
configuration_path = '../../../data/SearchEnginesCran.csv'
ground_truth_dict = utils.read_ground_truth(ground_truth_path)
search_engine_conf = utils.read_json(search_engine_conf_path)
MRR_results = MRR(ground_truth_dict, search_engine_conf)
R_precision_results = R_Precision(ground_truth_dict, search_engine_conf)
configurations = pd.read_csv(configuration_path)
configurations['MRR'] = MRR_results
configurations['Mean'] = np.mean(R_precision_results, axis=1)
configurations['Min'] = np.min(R_precision_results, axis=1)
configurations['Max'] = np.max(R_precision_results, axis=1)
configurations['Median'] = np.median(R_precision_results, axis=1)
configurations['1_quartile'] = np.quantile(a=R_precision_results,
q=.25,
axis=1)
configurations['3_quartile'] = np.quantile(a=R_precision_results,
q=.75,
axis=1)
configurations_top_5 = configurations.sort_values(
by=['MRR'], ascending=False).head(5).SE_ID
print(list(configurations_top_5))
col_names = ['Conf_' + str(i) for i in list(configurations_top_5)]
search_engine_conf_top_5 = {
key: search_engine_conf[key]
for key in configurations_top_5
}
print('P@k....')
P_at_k_res = P_at_k(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(P_at_k_res).transpose()
temp_df.columns = col_names
figure, axes = plt.subplots(1, 2, figsize=(8, 3))
ax1 = temp_df.plot(title='P@k Cranfield Data', ax=axes[0], style='.-')
ax1.set_xticks(positions)
ax1.set_xlabel('k values')
ax1.set_ylabel('Mean P@k')
print('NCDG@k....')
ncdg_at_k_res = ncdg(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(ncdg_at_k_res).transpose()
temp_df.columns = col_names
ax2 = temp_df.plot(title='NCDG@k Cranfield Data', ax=axes[1], style='.-')
ax2.set_xticks(positions)
ax2.set_xlabel('k values')
ax2.set_ylabel('Mean NCDG@k')
plt.tight_layout()
figure.savefig('../../../Report/Images/CranPlot.png')
configurations = configurations.sort_values(by=['MRR'], ascending=False)
configurations.to_csv(r'../../../data/SearchEnginesResultsCran.csv',
index=False)
# TIME
ground_truth_path = '../../../data/time_Ground_Truth.tsv'
search_engine_conf_path = '../../../data/total_query_results_time.json'
configuration_path = '../../../data/SearchEnginesTime.csv'
ground_truth_dict = utils.read_ground_truth(ground_truth_path)
search_engine_conf = utils.read_json(search_engine_conf_path)
MRR_results = MRR(ground_truth_dict, search_engine_conf)
R_precision_results = R_Precision(ground_truth_dict, search_engine_conf)
configurations = pd.read_csv(configuration_path)
configurations['MRR'] = MRR_results
configurations['Mean'] = np.mean(R_precision_results, axis=1)
configurations['Min'] = np.min(R_precision_results, axis=1)
configurations['Max'] = np.max(R_precision_results, axis=1)
configurations['Median'] = np.median(R_precision_results, axis=1)
configurations['1_quartile'] = np.quantile(a=R_precision_results,
q=.25,
axis=1)
configurations['3_quartile'] = np.quantile(a=R_precision_results,
q=.75,
axis=1)
configurations_top_5 = configurations.sort_values(
by=['MRR'], ascending=False).head(5).SE_ID
print(list(configurations_top_5))
col_names = ['Conf_' + str(i) for i in list(configurations_top_5)]
search_engine_conf_top_5 = {
key: search_engine_conf[key]
for key in configurations_top_5
}
print('P@k....')
P_at_k_res = P_at_k(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(P_at_k_res).transpose()
temp_df.columns = col_names
figure, axes = plt.subplots(1, 2, figsize=(8, 3))
ax1 = temp_df.plot(title='P@k Time Data', ax=axes[0], style='.-')
ax1.set_xticks(positions)
ax1.set_xlabel('k values')
ax1.set_ylabel('Mean P@k')
print('NCDG@k....')
ncdg_at_k_res = ncdg(ground_truth_dict, search_engine_conf_top_5)
temp_df = pd.DataFrame(ncdg_at_k_res).transpose()
temp_df.columns = col_names
ax2 = temp_df.plot(title='NCDG@k Time Data', ax=axes[1], style='.-')
ax2.set_xticks(positions)
ax2.set_xlabel('k values')
ax2.set_ylabel('Mean NCDG@k')
plt.tight_layout()
figure.savefig('../../../Report/Images/TimePlot.png')
configurations = configurations.sort_values(by=['MRR'], ascending=False)
configurations.to_csv(r'../../../data/SearchEnginesResultsTime.csv',
index=False)