def collect_svc(name, most_similar_name, perc):
svc_results = []
data = dataTools.Authorship(name, ratioTrain, ratioValid, dataPath)
logging.info('Training SVM on {0}'.format(name))
phi, indices = load_phi(name, data, percentage=perc)
if indices.shape[0] < 2:
return [{'acc': 0, 'f1': 0, 'auc': 0, 'prec': 0, 'time': 0}]
for split_n in range(NO_OF_SPLITS):
data.get_split(name, ratioTrain, ratioValid)
data.reduce_dim(indices)
start = time.perf_counter()
svc = linear_class.train_svc(data)
end = time.perf_counter()
svc_result = evaluate_svc(svc, data)
svc_result['time'] = start - end
svc_results.append(svc_result)
logging.info(
'SPLIT {0}: SVM results successfully collected: {1}'.format(
split_n, svc_result))
return svc_results
def main():
Path("results/one_vs_one").mkdir(parents=True, exist_ok=True)
logging.info('Loading data for One-vs-One.')
data_general = dataTools.Authorship('poe', 1, 0, dataPath)
logging.info('Computing Dissimilarty matrix')
dis_m = signal_similarity.get_dissimilarity_matrix(data=data_general)
logging.info('Starting training.')
svc_results, GCNN_results = check_prev_results()
for idx, name in enumerate(all_author_names):
most_similar_name = get_most_similar_name(dis_m, idx)
logging.info('Training GCNN and SVM on {0} v.s. {1}'.format(name, most_similar_name))
try:
if name not in svc_results.keys():
svc_results[name] = collect_svc(name, most_similar_name)
logging.info('SVM results successfully collected: {0}'.format(svc_results[name]))
if name not in GCNN_results.keys():
GCNN_results[name] = collect_gcnn(name, most_similar_name)
logging.info('GCNN results successfully collected: {0}'.format(GCNN_results[name]))
except BaseException as e:
dump_crash(GCNN_results, svc_results)
logging.error("Exception during collecting one-vs-one", exc_info=True)
raise e
dump_results(GCNN_results, svc_results)
def collect_gcnn(name, most_similar_name, perc):
gcnn_results = []
data = dataTools.Authorship(name, ratioTrain, ratioValid, dataPath)
logging.info('Training GCNN on {0}'.format(name))
h_params = ClusterUtils.load_best_hyperparams(name)
phi, indices = load_phi(name, data, percentage=perc)
if indices.shape[0] < 2:
return [{'acc': 0, 'f1': 0, 'auc': 0, 'prec': 0, 'time': 0}]
for split_n in range(NO_OF_SPLITS):
data.get_split(name, ratioTrain, ratioValid)
data.reduce_dim(indices)
start = time.perf_counter()
gcnn = train_helper.train_net(data, h_params, phi=phi)
end = time.perf_counter()
gcnn_eval = evaluate_gcnn(gcnn, data)
gcnn_eval['time'] = start - end
gcnn_results.append(gcnn_eval)
logging.info(
'SPLIT {0}: GCNN results successfully collected: {1}'.format(
split_n, gcnn_results[split_n]))
return gcnn_results
def main():
Path("results/one_vs_one").mkdir(parents=True, exist_ok=True)
logging.info('Loading data for One-vs-One.')
data_general = dataTools.Authorship('poe', 1, 0, dataPath)
logging.info('Computing Dissimilarty matrix')
dis_m = signal_similarity.get_dissimilarity_matrix(data=data_general)
logging.info('Starting training.')
GCNN_results = {}
svc_results = {}
for idx, name in enumerate(all_author_names):
most_similar_name = get_most_similar_name(dis_m, idx)
logging.info('Training GCNN and SVM on {0} v.s. {1}'.format(
name, most_similar_name))
GCNN_results[name] = {}
svc_results[name] = {}
try:
feature_search = [10, 15, 20, 25, 30, 35, 40, 45]
for feture_count in feature_search:
logging.info(
'Starting training with feature count: {0}'.format(
feture_count))
phi, indices = load_phi(name, no_of_features=feture_count)
svc_results[name][feture_count] = collect_svc(
name, most_similar_name, indices)
logging.info('SVM results successfully collected: {0}'.format(
svc_results[name][feture_count]))
GCNN_results[name][feture_count] = collect_gcnn(
name, most_similar_name, phi, indices)
logging.info('GCNN results successfully collected: {0}'.format(
GCNN_results[name][feture_count]))
dump_results(GCNN_results, svc_results)
except BaseException as e:
dump_crash(GCNN_results, svc_results)
logging.error(
"Exception during collecting one-vs-one with feature search",
exc_info=True)
raise e
def get_dissimilarity_matrix(data=None):
if data is None:
# the results from each run
dataDir = 'authorData' # Data directory
dataFilename = 'authorshipData.mat' # Data filename
dataPath = os.path.join(dataDir, dataFilename) # Data path
data = dataTools.Authorship('poe', 1, 0, dataPath=dataPath)
all_author_names = data.authorData.keys()
authors_mean_signals = []
for name in all_author_names:
curr_signals = data.authorData[name]["wordFreq"]
avg = get_author_average(curr_signals)
authors_mean_signals.append(avg)
dis_matrix = squareform(pdist(numpy.array(authors_mean_signals)))
return dis_matrix
def collect_svc(name, indices):
svc_results = []
data = dataTools.Authorship(name, ratioTrain, ratioValid, dataPath)
logging.info('Training SVM on {0}'.format(name))
for split_n in range(NO_OF_SPLITS):
data.get_split(name, ratioTrain, ratioValid)
data.reduce_dim(indices)
svc = linear_class.train_svc(data)
svc_result = evaluate_svc(svc, data)
svc_results.append(svc_result)
logging.info(
'SPLIT {0}: SVM results successfully collected: {1}'.format(
split_n, svc_result))
return svc_results
def collect_gcnn(name, phi, ind):
gcnn_results = []
data = dataTools.Authorship(name, ratioTrain, ratioValid, dataPath)
logging.info('Training GCNN on {0}'.format(name))
h_params = ClusterUtils.load_best_hyperparams(name)
for split_n in range(NO_OF_SPLITS):
data.get_split(name, ratioTrain, ratioValid)
data.reduce_dim(ind)
gcnn = train_helper.train_net(data, h_params, phi=phi)
gcnn_results.append(evaluate_gcnn(gcnn, data))
logging.info(
'SPLIT {0}: GCNN results successfully collected: {1}'.format(
split_n, gcnn_results[split_n]))
return gcnn_results
def main():
Path("results/one_vs_all").mkdir(parents=True, exist_ok=True)
logging.info('Loading data for One-vs-All with fingerprint matrix.')
data_general = dataTools.Authorship('poe', 1, 0, dataPath)
logging.info('Computing Dissimilarty matrix')
dis_m = signal_similarity.get_dissimilarity_matrix(data=data_general)
logging.info('Starting training.')
GCNN_results = {}
svc_results = {}
for idx, name in enumerate(all_author_names):
logging.info('Training GCNN and SVM on {0}'.format(name))
try:
phi, indices = load_phi(name)
svc_results[name] = collect_svc(name, indices)
logging.info('SVM results successfully collected: {0}'.format(
svc_results[name]))
GCNN_results[name] = collect_gcnn(name, phi, indices)
logging.info('GCNN results successfully collected: {0}'.format(
GCNN_results[name]))
except BaseException as e:
dump_crash(GCNN_results, svc_results)
logging.error("Exception during collecting one-vs-All",
exc_info=True)
raise e
dump_results(GCNN_results, svc_results)
def main():
Path("results/one_vs_one").mkdir(parents=True, exist_ok=True)
logging.info('Loading data for One-vs-One.')
data_general = dataTools.Authorship('poe', 1, 0, dataPath)
logging.info('Computing Dissimilarty matrix')
dis_m = signal_similarity.get_dissimilarity_matrix(data=data_general)
logging.info('Starting training.')
GCNN_results, svc_results = load_prev_results()
for idx, name in enumerate(all_author_names):
most_similar_name = get_most_similar_name(dis_m, idx)
logging.info('Training GCNN and SVM on {0} v.s. {1}'.format(
name, most_similar_name))
if name not in svc_results.keys():
svc_results[name] = {}
if name not in GCNN_results.keys():
GCNN_results[name] = {}
try:
percentages = [
0.01, 0.02, 0.05, .1, .15, .20, .25, .30, 0.35, 0.4, 0.5
]
# percentages = [0.05, .1, .15, .20, .25, .30]
for perc in percentages:
logging.info(
'Starting training with feature count: {0}'.format(perc))
if str(perc) in svc_results[name].keys() and len(
svc_results[name][str(perc)]) == 10:
logging.info('Skipping SVC for {1} with perc: {0}'.format(
perc, name))
else:
svc_results[name][str(perc)] = collect_svc(
name, most_similar_name, perc)
logging.info(
'SVM results successfully collected: {0}'.format(
svc_results[name][str(perc)]))
if str(perc) in GCNN_results[name].keys() and len(
GCNN_results[name][str(perc)]) == 10:
logging.info('Skipping GCNN for {1} with perc: {0}'.format(
perc, name))
else:
GCNN_results[name][str(perc)] = collect_gcnn(
name, most_similar_name, perc)
logging.info(
'GCNN results successfully collected: {0}'.format(
GCNN_results[name][str(perc)]))
dump_results(GCNN_results, svc_results)
except BaseException as e:
dump_crash(GCNN_results, svc_results)
logging.error(
"Exception during collecting one-vs-one with feature search",
exc_info=True)
raise e