def multiprocess_overlapping_cases(data: pd.DataFrame, job_num: int):
"""
Compute parallel executed case instances in given dataset. The
computation will be split into the given number of jobs.
:param data: case log
:param job_num: number of jobs
"""
print_time('parallel cases %s' % len(data))
# get split points
steps = get_steps(data, job_num)
print(steps)
jobs = []
out_q = Queue()
# start all jobs
for idx, r in enumerate(steps):
p = Process(target=overlapping_cases, args=(data, idx + 1, r, out_q))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
# collect processes
for job in jobs:
job.join()
# update case log
for k, v in res.items():
data.at[k, 'overlapping_cases'] = v
def _optimize_model(self, arg_values : argparse.Namespace) -> \
Iterable[GoalEncState]:
with print_time("Loading data", guard=arg_values.verbose):
if arg_values.start_from:
_, (arg_values, unparsed_args, (metadata, state)) = \
torch.load(arg_values.start_from)
_, tokenized_goals, outputs = \
goals_to_total_distances_tensors_with_meta(
extract_dataloader_args(arg_values),
str(arg_values.scrape_file), metadata)
else:
metadata, tokenized_goals, outputs = \
goals_to_total_distances_tensors(
extract_dataloader_args(arg_values),
str(arg_values.scrape_file))
with print_time("Converting data to tensors", guard=arg_values.verbose):
tensors = [pad_sequence([torch.LongTensor(tok_goal)
for tok_goal in tokenized_goals],
batch_first=True),
torch.FloatTensor(outputs)]
with print_time("Building the model", guard=arg_values.verbose):
model = self._get_model(arg_values, goal_enc_get_num_tokens(metadata))
if arg_values.start_from:
self.load_saved_state(arg_values, unparsed_args, state)
return ((metadata, state) for state in
optimize_checkpoints(tensors, arg_values, model,
lambda batch_tensors, model:
self._get_batch_prediction_loss(arg_values,
batch_tensors,
model)))
def parallel_activities(data: pd.DataFrame, delta: datetime.timedelta,
instance, range: list, out_q: Queue):
"""
This function is called by multiprocess_parallel_activities(). It computes
the number of parallel executed activity instances in a given subset.
:param data: dataset
:param delta: time interval
:param instance: current instance
:param range: subset
:param out_q: output queue
"""
# total number of activity instances
size = len(data)
# collect results
r = {}
for idx, row in data[range[0]:range[1]].iterrows():
exclude_cases = data.loc[(data['start'] > row['end'] + delta) |
(data['end'] < row['start'] - delta)]
counter = size - len(exclude_cases)
r[idx] = counter
print_time('instance: %s' % instance, False)
out_q.put(r)
def get_tokens(args: List[str]):
parser = argparse.ArgumentParser(description="Pick a set of tokens")
parser.add_argument("--type", choices=["mixed"], default="mixed")
parser.add_argument("-v", "--verbose", action='count', default=0)
parser.add_argument("-n", "--num-keywords", type=int, default=120)
parser.add_argument("-s", "--num-samples", type=int, default=2000)
parser.add_argument("-j", "--num-threads", type=int, default=None)
parser.add_argument("scrapefile", type=Path2)
parser.add_argument("dest")
arg_values = parser.parse_args(args)
with print_time("Reading scraped data", guard=arg_values.verbose):
raw_data = list(data.read_text_data(arg_values.scrapefile))
embedding = SimpleEmbedding()
subset = data.RawDataset(random.sample(raw_data, arg_values.num_samples))
relevance_pairs = [
(context.focused_goal,
embedding.encode_token(serapi_instance.get_stem(tactic)))
for relevant_lemmas, prev_tactics, context, tactic in subset
]
with print_time("Calculating keywords", guard=arg_values.verbose):
keywords = get_relevant_k_keywords2(relevance_pairs,
arg_values.num_keywords,
arg_values.num_threads)
with (open(arg_values.dest, mode='w') if arg_values.dest != "-" else
contextlib.nullcontext(sys.stdout)) as f:
for keyword in keywords:
f.write(keyword + "\n")
def train(self, checkpoint_path=None, weights_only=False):
print('Starting training')
print(datetime.datetime.now())
start_time = datetime.datetime.now()
# Load pretrained parameters if desired
if checkpoint_path is not None:
self.load_checkpoint(checkpoint_path, weights_only)
if weights_only:
self.initialize_visualizations()
else:
# Initialize any training visualizations
self.initialize_visualizations()
# Train for specified number of epochs
for self.epoch in range(self.epoch, self.num_epochs):
epoch_start_time = datetime.datetime.now()
# Increment the LR scheduler
if self.scheduler is not None:
self.scheduler.step()
# Run an epoch of training
self.train_one_epoch()
epoch_end_time = datetime.datetime.now()
total_seconds = (epoch_end_time - epoch_start_time).seconds
util.print_time('Epoch', total_seconds)
if self.epoch % self.validation_freq == 0:
self.validate()
if self.lin_rms_sq_error_meter.avg <= self.lin_rms_sq_error and self.loss.avg <= self.loss_error:
self.save_checkpoint()
self.lin_rms_sq_error = self.lin_rms_sq_error_meter.avg
self.loss_error = self.loss.avg
self.visualize_metrics()
end_time = datetime.datetime.now()
seconds = (end_time - start_time).seconds
util.print_time('Training', seconds)
def set_activity_instances(df: pd.DataFrame):
"""
This function maps events and activity instances based on a first-come,
first-serves approach.
:param df: event log
"""
print_time('set activity instance')
end_transitions = [
'autoskip', 'manualskip', 'complete', 'withdraw', 'ate_abort',
'pi_abort'
]
trace = df.loc[0, 'caseID']
activity = 0
instances = {}
for idx, row in df.iterrows():
if row['caseID'] != trace:
trace = row['caseID']
instances = {}
if row['name'] not in instances.keys():
activity += 1
instances[row['name']] = activity
df.at[idx, 'activity_instance'] = instances[row['name']]
if row['transition'].lower() in end_transitions:
instances.pop(row['name'], None)
print_time('set activity instance', False)
def import_report(file: str):
"""
This function is used to import a conformance report.
:param file: conformance report
:return: pandas DataFrame, containing raw cost per case
"""
print_time('Import report')
df = pd.read_csv(file, sep=',', header=[0])
costs = []
cases = []
for idx, row in enumerate(df.loc[:, 'Case IDs']):
raw_cost = df.loc[idx, 'Raw Fitness Cost']
# update alignment id in case table
for case in df.loc[idx, 'Case IDs'].split('|'):
cases.append(case)
costs.append(float(raw_cost))
print_time('Import report', False)
df = pd.DataFrame({'case': cases, 'response': costs})
return df
def verify_candidates(candidates, user_movies_matrix, start_time):
print("\nVerifying candidates...")
count = 0
print("Number of buckets in total: " + str(len(candidates)))
for cnr, candidate_group in enumerate(candidates):
# print("Number of candidates in bucket " + str(cnr) + ": " + str(len(candidate_group)))
for cnr1, candidate1 in enumerate(candidate_group):
for cnr2 in range(cnr1 + 1, len(candidate_group)):
candidate2 = list(candidate_group)[cnr2]
jsim = sim.jaccard(user_movies_matrix[candidate1],
user_movies_matrix[candidate2])
if jsim >= 0.50:
print("Number of candidates in bucket " + str(cnr) + ": " +
str(len(candidate_group)))
count = count + 1
print((candidate1, candidate2))
print("Similarity: " + str(
sim.jaccard(user_movies_matrix[candidate1],
user_movies_matrix[candidate2])))
print("Found until now: " + str(count))
util.print_time(start_time)
print()
# print()
print(count)
def open_page(self, url, wait_time=0):
print_time(f"Opening url {url}")
self.driver.get(url)
if (wait_time > 0):
self.driver.implicitly_wait(wait_time)
print_time("Page loaded")
def multiprocess_overlapping_events(log_data: pd.DataFrame, job_num: int):
"""
Compute parallel executed events of given dataset. The computation will be
split into the given number of jobs.
:param log_data: event log
:param job_num: number of jobs
"""
print_time('parallel events')
# get all event names
event_names = log_data['name'].unique()
# convert variable type of timestamp
log_data['timestamp'] = pd.to_datetime(log_data.loc[:, 'timestamp'],
format='%Y-%m-%d %H:%M:%S')
# iterate through all event names
for e in event_names:
print_time('calculate parallel events for %s' % e)
# get subset
sub_data = log_data.loc[log_data['name'] == e]
sub_data = sub_data.sort_values(by=['timestamp'])
steps = get_steps(sub_data, job_num)
print(steps)
# collect jobs and results
jobs = []
out_q = Queue()
# set time interval (= theta)
delta = datetime.timedelta(days=1)
# start different jobs
for idx, r in enumerate(steps):
p = Process(target=overlapping_events,
args=(sub_data, delta, idx + 1, r, out_q))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
# collect processes
for job in jobs:
job.join()
for k, v in res.items():
log_data.at[k, 'parallel_events'] = v
def multiprocess_parallel_event_sets(data: pd.DataFrame, job_num: int):
"""
Compute parallel executed event sets of given dataset. The computation
will be split into the given number of jobs.
:param data: event log
:param job_num: number of jobs
"""
print_time('parallel event sets')
data['parallel_sets'] = [0] * len(data)
set_names = data['set_name'].unique()
print('Set names: %s' % set_names)
# iterate through sets
for s in set_names:
print_time('calculate parallel events sets for set_%s' % s)
sub_data = data.loc[data['set_name'] == s]
# get split points
steps = get_steps(sub_data, job_num)
print(steps)
# collect jobs and results
jobs = []
out_q = Queue()
# set time interval (= theta)
delta = datetime.timedelta(days=1)
# start jobs
for idx, r in enumerate(steps):
p = Process(target=parallel_event_sets, args=(
sub_data, delta,
idx + 1,
r, out_q))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
# collect processes
for job in jobs:
job.join()
# update data
for k, v in res.items():
data.at[k, 'parallel_sets'] = v
def build_tag_to_pos(Y):
tag_to_pos = {}
i = 0
print_time("building build_tag_to_pos...")
for s in Y:
for t in s:
if t not in tag_to_pos:
tag_to_pos[t] = i
i += 1
pos_to_tag = {v: k for k, v in tag_to_pos.items()}
return tag_to_pos, pos_to_tag
def reinforce_training_worker(args: argparse.Namespace,
initial_buffer_size: int,
lock: Lock,
namespace: multiprocessing.managers.Namespace,
samples: Queue[LabeledTransition]):
last_trained_at = 0
samples_retrieved = 0
memory: List[LabeledTransition] = []
while True:
if samples_retrieved - last_trained_at < args.train_every_min:
next_sample = samples.get()
memory.append(next_sample)
samples_retrieved += 1
continue
else:
try:
next_sample = samples.get(timeout=.01)
memory.append(next_sample)
samples_retrieved += 1
if samples_retrieved - last_trained_at > args.train_every_max:
eprint("Forcing training", guard=args.verbose >= 2)
else:
continue
except queue.Empty:
pass
if len(memory) > args.buffer_max_size:
memory = random.sample(memory, args.buffer_max_size -
args.train_every_max)
# del memory[0:args.train_every_max+1]
if samples_retrieved - last_trained_at >= args.train_every_min:
last_trained_at = samples_retrieved
transition_samples = sample_batch(memory, args.batch_size)
with lock:
eprint(
f"Locked in training thread for {len(memory)} samples",
guard=args.verbose >= 2)
q_estimator = namespace.estimator
predictor = namespace.predictor
with print_time("Assigning scores", guard=args.verbose >= 2):
training_samples = assign_scores(args,
q_estimator,
predictor,
transition_samples)
with print_time("Training", guard=args.verbose >= 2):
q_estimator.train(training_samples,
show_loss=args.show_loss)
q_estimator.save_weights(args.out_weights, args)
namespace.estimator = q_estimator
eprint("Unlocked in training thread",
guard=args.verbose >= 2)
pass
def build_word_to_pos(X, padding="EOC"):
word_to_pos = {}
word_to_pos[padding] = 0
i = 1
print_time("building build_word_to_pos...")
for s in X:
for w in s:
if w not in word_to_pos:
word_to_pos[w] = i
i += 1
pos_to_word = {v: k for k, v in word_to_pos.items()}
return word_to_pos, pos_to_word
def fit(self, dataset, num_epochs, tolerance=0.0001):
"""
Parameters
----------
dataset:
Dataset with the sequences and tags
num_epochs: int
Number of epochs that the model will be trained
Returns
--------
Nothing. The method only changes self.parameters.
"""
print_time("Starting training...")
self.tolerance = tolerance
if self.fitted:
print("\n\tWarning: Model already trained")
if len(self.acc_per_epoch) == 0:
prev_acc = 0
else:
prev_acc = self.acc_per_epoch[-1]
for epoch in range(num_epochs):
acc = self.fit_epoch(dataset)
print_time("Epoch: %i Accuracy: %f" % (epoch, acc))
self.acc_per_epoch.append(acc)
if abs(acc-prev_acc) < self.tolerance:
print("Stopped by tolerance!")
break
prev_acc = acc
if self.averaged:
new_w = 0
for old_w in self.params_per_epoch:
new_w += old_w
new_w /= len(self.params_per_epoch)
self.parameters = new_w
self.fitted = True
def multiprocess_parallel_activities(data: pd.DataFrame, job_num: int):
"""
Compute parallel executed activity instances in given dataset. The
computation will be split into the given number of jobs.
:param data: activity log
:param job_num: number of jobs
"""
print_time('parallel activities')
activity_name = data['name'].unique()
# compute for all activities
for a in activity_name:
print_time('calculate parallel activities for activity %s' % a)
# get subset
sub_data = data.loc[data['name'] == a]
steps = get_steps(sub_data, job_num)
print(steps)
jobs = []
out_q = Queue()
# set time interval (=theta)
delta = datetime.timedelta(days=1)
# start all jobs
for idx, r in enumerate(steps):
p = Process(target=parallel_activities,
args=(sub_data, delta, idx + 1, r, out_q))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
# collect processes
for job in jobs:
job.join()
# update DataFrame
for k, v in res.items():
data.at[k, 'parallel_activities'] = v
def print_report(nr_found, nr_buckets, start_time, matrix_shape, k, b, r,
seed):
print("______________________________________________\n")
print("Final Report")
print("------------\n")
print("Found pairs in total: " + str(nr_found))
print("Number of buckets in total: " + str(nr_buckets))
util.print_time(start_time)
print()
print("For:")
print('Users: {1} | Movies: {0}'.format(str(matrix_shape[0]),
str(matrix_shape[1])))
print()
print("With parameters:")
print('Sig. Length: {0} | Bands: {1} | Rows: {2} | Seed: {3}'.format(
k, b, r, seed))
def callback(self):
print('==')
global first
if self.img_ready():
with print_time('get_img'):
surface, img_file = self.get_img()
with print_time('gui'):
self.set_display(surface)
self.draw_line((self.xmid_px, 0),
(self.xmid_px, self.height_px))
if do_face:
with print_time('get_face'):
result = self.get_face(img_file)
if result:
(mouth_x, mouth_y), depth = result
if first:
print(
f'Field of view: {self.width_px}x{self.height_px} px'
)
print(
f'Field of view: {self.width_deg:.0f}x{self.height_deg:.0f} deg'
)
print(
f'degrees: {self.deg_per_px_x:.3f} {self.deg_per_px_y:.3f}'
)
first = False
self.draw_dot((mouth_x, mouth_y))
with print_time('do_servo'):
if do_servo:
altitude = (self.ymid_px -
mouth_y) * self.deg_per_px_y
self.turn(self.xmid_px - mouth_x)
self.aim(altitude, depth)
self.got_target()
if do_shoot:
self.maybe_fire()
else:
self.cancel_target()
else:
print('image not ready')
def extract_data_from_table(self, template, pop=[], remove_last=False):
if (not template):
raise Exception("No template added to parser")
print_time("Extracting data from table")
data = []
rows = self.soup.find_all("tr")
if (len(pop) > 0):
sorted_pop = pop.sort(reverse=True)
for index in sorted_pop:
print_time(f"Removing row by index {index}")
rows.pop(index) # Remove table header
if (remove_last):
print_time(f"Removing last element")
rows.pop()
current = 1
log_counter = 0
elements = len(rows)
#Iterate over every row in the table
for row in rows:
log_counter += 1
if (log_counter == self.log_each_n or current == elements):
print_time(f"Parsing row {current} of {elements}")
log_counter = 0
new_dict = {}
columns = row.find_all("td")
if (type(template) == "dict"):
for key, value in template.items():
new_dict[value] = columns[key].text
else:
for index, value in enumerate(template):
new_dict[value] = columns[index].text
data.append(new_dict)
current += 1
print_time("Parser complete")
return data
def _optimize_model(
self, arg_values: argparse.Namespace
) -> Iterable[FeaturesDNNEvaluatorState]:
with print_time("Loading data", guard=arg_values.verbose):
if arg_values.start_from:
_, (arg_values, unparsed_args,
(picklable_token_map,
state)) = torch.load(arg_values.start_from)
token_map = tmap_from_picklable(picklable_token_map)
_, word_features_data, vec_features_data, outputs,\
word_features_vocab_sizes, vec_features_size = features_to_total_distances_tensors_with_map(
extract_dataloader_args(arg_values),
str(arg_values.scrape_file), token_map)
else:
token_map, word_features_data, vec_features_data, outputs, \
word_features_vocab_sizes, vec_features_size = features_to_total_distances_tensors(
extract_dataloader_args(arg_values),
str(arg_values.scrape_file))
# eprint(f"word data: {word_features_data[:10]}")
# eprint(f"vec data: {vec_features_data[:10]}")
# eprint(f"outputs: {outputs[:100]}")
with print_time("Converting data to tensors",
guard=arg_values.verbose):
tensors = [
torch.LongTensor(word_features_data),
torch.FloatTensor(vec_features_data),
torch.FloatTensor(outputs)
]
with print_time("Building the model", guard=arg_values.verbose):
model = self._get_model(arg_values, word_features_vocab_sizes,
vec_features_size)
if arg_values.start_from:
self.load_saved_state(arg_values, unparsed_args, state)
return (
(tmap_to_picklable(token_map), state)
for state in optimize_checkpoints(
tensors, arg_values, model, lambda batch_tensors, model: self.
_get_batch_prediction_loss(arg_values, batch_tensors, model)))
def mark_outlier(data: pd.DataFrame, base_path: str):
"""
This function calls the LOF computation for each feature.
:param data: case log
:param base_path: path to save plots
"""
print_time('events: mark outlier')
columns = []
for c in list(data.columns):
# ignore selected columns
if 'case' in c or 'involved' in c:
pass
else:
columns.append(c)
# call LOF computation
for c in columns:
do_LOF(data, c, base_path, plot=False)
def mark_outlier(data: pd.DataFrame, base_path: str):
"""
This function calls the LOF computation for each feature.
:param data: case log
:param base_path: path to location to save plots
"""
print_time('case mark outlier')
columns = []
for c in list(data.columns):
# ignore selected columns
if 'case_duration' in c or 'overlapping_cases' in c:
columns.append(c)
else:
pass
# call LOF computation
for c in columns:
do_LOF(data, c, base_path, plot=True)
def verify_partial_candidates(candidate_group, user_movies_matrix, bucket_nr,
nr_found, start_time):
for cnr1, candidate1 in enumerate(candidate_group):
for cnr2 in range(cnr1 + 1, len(candidate_group)):
candidate2 = list(candidate_group)[cnr2]
jsim = sim.jaccard(user_movies_matrix[candidate1],
user_movies_matrix[candidate2])
if jsim >= 0.50:
pair = sorted((candidate1, candidate2))
data.save_pair(pair)
print("\tFound similar pair: " + str(pair))
print("\tSimilarity: " + str(
sim.jaccard(user_movies_matrix[candidate1],
user_movies_matrix[candidate2])))
print("\tBucket number: " + str(bucket_nr))
print("\tNumber of candidates in the bucket: " +
str(len(candidate_group)))
nr_found[0] = nr_found[0] + 1
print("\tFound until now: " + str(nr_found[0]))
util.print_time(start_time, "\t")
print()
def find_cluster(data: pd.DataFrame, base_path: str, plot: bool):
"""
This function calls the k-means clustering for all features.
:param data: case log
:param base_path: path to save plots and cluster information
:param plot: whether to generate plots or not
"""
print_time('case kmeans')
columns = []
for c in list(data.columns):
# ignore selected columns
if 'case_duration' in c or 'overlapping_cases' in c:
columns.append(c)
else:
pass
# call k-means computation
for c in columns:
do_kmeans(data, c, base_path, plot)
def find_cluster(data: pd.DataFrame, base_path: str, plot: bool):
"""
This function calls the k-means clustering for all features.
:param data: case log
:param base_path: path to save plots and cluster information
:param plot: whether to generate plots or not
"""
print_time('activities kmeans')
columns = []
for c in list(data.columns):
# ignore selected columns
if 'case' in c or 'involved' in c or 'weekday' in c or 'weekend' in c \
or 'start_am' in c:
pass
else:
columns.append(c)
# call k-means computation
for c in columns:
do_kmeans(data, c, base_path, plot)
def run(bolt_path, plink_path, bfile, num_people, pheno_path, pheno_col, out_id):
info = (f'=> generating subset individual data\n'
f'bolt_path: {bolt_path}\n'
f'plink_path: {plink_path}\n'
f'bfile: {bfile}\n'
f'num_people: {num_people}\n'
f'pheno_filename: {pheno_path}\n'
f'pheno_col: {pheno_col}\n'
f'out: {out_id}\n')
print(info)
sys.stdout.flush()
file_cache_out_path = os.path.join('file_cache', f'{out_id}_{num_people}')
_, pheno_temp = generate_subset(plink_path=plink_path, bfile=bfile, num_people=num_people, out=file_cache_out_path,
pheno_path=pheno_path)
print(f'subset pheno file: {pheno_temp}')
print('=> assigning the SNP components by chromosome')
sys.stdout.flush()
snp_assignment_filename = file_cache_out_path + '.snps_assignment'
partition(file_cache_out_path + '.bim', snp_assignment_filename)
print('=> running BOLT-REML')
print_time()
sys.stdout.flush()
dt = bench_bolt_reml(bolt_path, snp_assignment_filename,
file_cache_out_path + '.bed',
file_cache_out_path + '.bim',
file_cache_out_path + '.fam',
pheno_temp, pheno_col)
log_path_prefix = os.path.join('output', f'{out_id}_{num_people}')
print(f'log_path_prefix: {log_path_prefix}')
with open(f'{log_path_prefix}.bench', 'w') as file:
file.write(info)
file.write(f'BOLT-REML took {dt} sec\n')
print_time()
def test():
if(len(sys.argv) < 3):
raise Exception("Script must be called with two arguments, the path to chromedriver and the path to firebase config")
chromedriver = sys.argv[1]
elapsed = Elapsed()
scraper = Scraper(chromedriver, headless=True)
test_url = "https://96hpr.csb.app"
try:
scraper.open_page(test_url)
html = scraper.get_outerhtml(
By.XPATH, "/html/body/div/div/table/tbody")
parsed = Parser(html, log_each_n=10)
template = ["A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M", "N", "O"]
parsed.extract_data_from_table(template, [0], True)
print_time(f"Extracted data")
finally:
scraper.close()
elapsed.end()
def multiprocess_aggregation(data: pd.DataFrame, job_num: int):
"""
Aggregate enriched event log to case log. The computation will be
split into the given number of jobs.
:param data: event log
:param job_num: number of jobs
:return: case log
"""
print_time('aggregate events')
print('Lenght dataset %s' % len(data))
case_ids = data['caseID'].unique()
steps = get_steps_seq(case_ids, data, job_num)
print(steps)
events = data['name'].unique()
# feature types
feature = [
'abs_lag', 'pre_lag', 'post_lag', 'weekday', 'weekend',
'parallel_events'
]
# blueprint to collect single events
case_tmp = {}
for e in events:
case_tmp[e] = {'counter': 0}
for f in feature:
case_tmp[e][f] = 0
jobs = []
out_q = Queue()
# start all jobs
for idx, r in enumerate(steps):
p = Process(target=aggregate, args=(data, case_tmp, r, out_q, idx))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
for job in jobs:
job.join()
print_time('time features', start=False)
print_time('merge results')
# return case log
return pd.DataFrame(list(res.values()))
def multiprocess_aggregation(data: pd.DataFrame, job_num: int):
"""
Aggregate enriched activity log to case log. The computation will be
split into the given number of jobs.
:param data: enriched activity log
:param job_num: number of jobs
:return: case log
"""
print_time('aggregate activities')
case_ids = data['caseID'].unique()
steps = get_steps_seq(case_ids, data, job_num)
# prepare names
activity_names = data['name'].unique()
activity_names = ['_'.join(['act', str(i)]) for i in activity_names]
feature = [
'abs_lag', 'duration', 'start_am', 'weekday', 'parallel_activities'
]
# generate blueprint
case_tmp = {}
for a in activity_names:
case_tmp[a] = {'counter': 0}
for f in feature:
case_tmp[a][f] = 0
jobs = []
out_q = Queue()
# start all jobs
for idx, r in enumerate(steps):
p = Process(target=aggregate, args=(data, case_tmp, r, out_q, idx))
jobs.append(p)
p.start()
# merge jobs and collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
for job in jobs:
job.join()
print_time('aggregate activities', start=False)
print_time('merge results')
return pd.DataFrame(list(res.values()))
def multiprocess_time_feature(log_data: pd.DataFrame, job_num: int):
"""
Compute basic features for given dataset. The computation will be split
into the given number of jobs.
:param log_data: event log
:param job_num: number of jobs
"""
print_time('time features')
case_ids = log_data['caseID'].unique()
# convert variable type of timestamp
log_data['timestamp'] = pd.to_datetime(log_data.loc[:, 'timestamp'],
utc=True,
format='%Y-%m-%d '
'%H:%M:%S')
# get split points
steps = get_steps_seq(case_ids, log_data, job_num)
jobs = []
out_q = Queue()
# start jobs
for idx, r in enumerate(steps):
p = Process(target=multi_time_feature, args=(log_data, r, out_q, idx))
jobs.append(p)
p.start()
# collect results
res = {}
for i in range(len(steps)):
res.update(out_q.get())
for job in jobs:
job.join()
print_time('time features', start=False)
print_time('merge results')
# add features to DataFrame
for idx, data in res.items():
for attr, v in data.items():
log_data.at[idx, attr] = v
from util import print_time
def main():
parser = argparse.ArgumentParser(description="Merge multiple svm format features.")
parser.add_argument(
"-i", nargs="*", required=True, dest="input_filename", help="Specify input file path (accept multiple inputs)"
)
parser.add_argument("-o", required=True, dest="output_filename", help="Specify output file path")
opts = parser.parse_args(sys.argv[1:])
all_X = []
for fileName in opts.input_filename:
print "Loading " + fileName + " ..."
X, y = load_svmlight_file(fileName)
print X.shape
all_X.append(X.todense())
X = np.concatenate(all_X, axis=1)
print "Saving " + opts.output_filename + " ..."
print X.shape
dump_svmlight_file(X, y, opts.output_filename)
if __name__ == "__main__":
ts = time.time()
main()
te = time.time()
print_time(ts, te)
psi_ez_x[-1,:-1,:] = -pml_cb[-1]*(hy[-1,:-1,:] - hy[-3,:-1,:])
for i in xrange(-2,-npml-1,-1):
psi_ez_x[i,:-1,:] = pml_ca[i]*psi_ez_x[i+1,:-1,:] - pml_cb[i]*(hy[i,:-1,:] - hy[i-2,:-1,:])
ez[-npml-2:-2,:-1,:] += 0.5*psi_ez_x[:,:-1,:]
# for source
ez[270,ny/2,1] += np.sin(2*np.pi*frequency*dt*tstep)
# for pbc
update_pbc_e('z', *em_arrays[:-3])
update_h(*em_arrays)
# for pml
psi_hy_x[-1,:,1:] = -pml_cb[-1]*(ez[-1,:,1:] - ez[-3,:,1:])
for i in xrange(-2,-npml-1,-1):
psi_hy_x[i,:,1:] = pml_ca[i]*psi_hy_x[i+1,:,1:] - pml_cb[i]*(ez[i,:,1:] - ez[i-2,:,1:])
hy[-npml-1:-1,:,1:] += 0.5*psi_hy_x[:,:,1:]
# for pbc
update_pbc_h('z', *em_arrays[3:])
if tstep%tgap == 0:
print_time(tstep)
#print hx[260:280,190:210,1]
#print psi_ez_x[:,ny/2,nz/2]
im.set_array(ez[:,:,nz/2].T**2)
plt.draw()
#savefig('./png/%.5d.png' % tstep)
print ''
