def test_candidates():
mm_engine = ex.create_mm_engine(openslex_file_path)
rs = cn.get_relationships(mm_engine)
assert rs.__len__() == 8
for r in rs:
assert r['rs'] in [
'CONCERT_HALL_FK', 'SEAT_HALL_FK', 'BP_CONCERT_FK', 'BP_BAND_FK',
'TICKET_CONCERT', 'TICKET_SEAT', 'BOOKING_CON', 'BOOKING_FK'
]
stats = cn.get_stats_mm(mm_engine)
os.makedirs('output/dumps/', exist_ok=True)
json.dump(stats, open('output/dumps/stats_mm.json', 'wt'), indent=True)
candidates = cn.compute_candidates(mm_engine)
# json.dump(candidates, open('output/dumps/candidates.json', 'wt'), indent=True)
random.seed(0)
random.shuffle(candidates)
for idx, cand in enumerate(
list(candidates.values())[:min(5, candidates.__len__())]):
log_name = 'log_test_{}'.format(idx)
print('Computing Log: {}'.format(log_name))
cn.build_log_for_case_notion(mm_engine,
cand,
proc_name='proc_test_{}'.format(idx),
log_name=log_name)
assert candidates.__len__() > 0
def check_mm(openslex_file_path, connection_params, metadata):
mm_engine = ex.create_mm_engine(openslex_file_path)
mm_conn = mm_engine.connect()
db_engine = ex.create_db_engine(**connection_params)
db_conn = db_engine.connect()
check = True
for t in metadata.tables.keys():
mm_q = sq.select([text('count(O.id) as num')]).select_from(text('object as O, class as CL')) \
.where(text('O.class_id == CL.id and CL.name == "{}"'.format(t)))
mm_res = mm_conn.execute(mm_q)
mm_num = mm_res.first()
mm_res.close()
db_q = sq.select([text('count(*) as num')]).select_from(text(t))
db_res = db_conn.execute(db_q)
db_num = db_res.first()
db_res.close()
check_t = mm_num['num'] == db_num['num']
check = check and check_t
assert check
mm_conn.close()
mm_engine.dispose()
db_conn.close()
db_engine.dispose()
print(check)
return check
def test_default_model(openslex=train_openslex_file_path,
ground_truth=ground_truth_path):
mm_engine = ex.create_mm_engine(openslex)
mm_meta = ex.get_mm_meta(mm_engine)
disc = ev.discover_event_definitions(mm_engine, mm_meta, model='default')
aid = disc['aid']
y_true_in_table = aid.load_y_true(disc[ev.CT_IN_TABLE]['candidates'],
ground_truth)
y_true_lookup = aid.load_y_true(disc[ev.CT_LOOKUP]['candidates'],
ground_truth)
y_true_ts_fields = aid.load_y_true(disc[ev.CT_TS_FIELD]['candidates'],
ground_truth)
scores_ts_fields = aid.score(y_true_ts_fields,
disc[ev.CT_TS_FIELD]['predicted'])
scores_in_table = aid.score(y_true_in_table,
disc[ev.CT_IN_TABLE]['predicted'])
scores_lookup = aid.score(y_true_lookup, disc[ev.CT_LOOKUP]['predicted'])
print('Score Ts Fields')
pprint(scores_ts_fields)
print('Score In Table')
pprint(scores_in_table)
print('Score Lookup')
pprint(scores_lookup)
def export_log(mm_path, log_id, output_file):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
df: pd.DataFrame = cn.log_to_dataframe(mm_engine=mm_engine,
mm_meta=mm_meta,
log_id=log_id)
df.to_csv(output_file, index_label='idx')
def print_cn(mm_path, cn_id, cn_dir, output_file, view):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
try:
candidates = pickle.load(open(Path(cn_dir, 'candidates.pkl'), 'rb'))
case_notion = candidates[cn_id]
dump_cn_dot(mm_engine, mm_meta, case_notion, output_file, view)
except:
raise Exception('No case notion to show')
def print_cn_log(mm_path, log_id, output_file, view):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
info = cn.log_info(mm_engine, mm_meta, log_id)
if 'case_notion' in info['attributes']:
dump_cn_dot(mm_engine, mm_meta, info['attributes']['case_notion'],
output_file, view)
else:
raise Exception('No case notion to show')
def test_candidates():
mm_engine = ex.create_mm_engine(train_openslex_file_path)
mm_meta = ex.get_mm_meta(mm_engine)
os.makedirs(dumps_dir, exist_ok=True)
classes = None
model = ev.train_model(mm_engine=mm_engine,
mm_meta=mm_meta,
y_true_path=ground_truth_path,
classes=classes,
model_output=trained_model)
disc = ev.discover_event_definitions(mm_engine=mm_engine,
mm_meta=mm_meta,
classes=classes,
dump_dir=dumps_dir,
model=model)
shutil.copyfile(train_openslex_file_path, modified_mm_path)
mm_engine_modif = ex.create_mm_engine(modified_mm_path)
mm_meta_modif = ex.get_mm_meta(mm_engine_modif)
pred_ts_cand = [
c for p, c in zip(disc[ev.CT_TS_FIELD]['predicted'], disc[
ev.CT_TS_FIELD]['candidates']) if p == 1
]
pred_in_table_cand = [
c for p, c in zip(disc[ev.CT_IN_TABLE]['predicted'], disc[
ev.CT_IN_TABLE]['candidates']) if p == 1
]
pred_lookup_cand = [
c for p, c in zip(disc[ev.CT_LOOKUP]['predicted'], disc[ev.CT_LOOKUP]
['candidates']) if p == 1
]
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_ts_cand[:2])
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_in_table_cand[:2])
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_lookup_cand[:2])
def disc_and_build(mm: Path, new_mm: Path, dump_dir: Path, build_events=False):
print("Discovering and building events for: {}".format(mm))
print("In: {}".format(new_mm))
print("Dumping in: {}".format(dump_dir))
mm_engine_train = ex.create_mm_engine(mm)
mm_meta_train = ex.get_mm_meta(mm_engine_train)
cached_dir_train = dump_dir
ts_train_path = cached_dir_train / 'timestamps.json'
candidates_ts_fields_path = cached_dir_train / 'candidates_ts_fields.json'
candidates_in_table_path = cached_dir_train / 'candidates_in_table.json'
candidates_lookup_path = cached_dir_train / 'candidates_lookup.json'
features_in_table_path = cached_dir_train / 'features_in_table.json'
features_lookup_path = cached_dir_train / 'features_lookup.json'
final_candidates_ts_fields_path = cached_dir_train / 'final_candidates_ts_fields.json'
final_candidates_in_table_path = cached_dir_train / 'final_candidates_in_table.json'
final_candidates_lookup_path = cached_dir_train / 'final_candidates_lookup.json'
os.makedirs(cached_dir_train, exist_ok=True)
aid = ev.ActivityIdentifierDiscoverer(engine=mm_engine_train,
meta=mm_meta_train,
model='default')
if os.path.exists(ts_train_path):
timestamp_attrs = aid.load_timestamp_attributes(ts_train_path)
else:
timestamp_attrs = aid.get_timestamp_attributes()
aid.save_timestamp_attributes(timestamp_attrs, ts_train_path)
#
if os.path.exists(candidates_ts_fields_path):
candidates_ts_fields = aid.load_candidates(candidates_ts_fields_path)
else:
candidates_ts_fields = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_TS_FIELD)
aid.save_candidates(candidates_ts_fields, candidates_ts_fields_path)
if os.path.exists(candidates_in_table_path):
candidates_in_table = aid.load_candidates(candidates_in_table_path)
else:
candidates_in_table = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_IN_TABLE)
aid.save_candidates(candidates_in_table, candidates_in_table_path)
if os.path.exists(candidates_lookup_path):
candidates_lookup = aid.load_candidates(candidates_lookup_path)
else:
candidates_lookup = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_LOOKUP)
aid.save_candidates(candidates_lookup, candidates_lookup_path)
#
if os.path.exists(features_in_table_path):
X_in_table = aid.load_features(features_in_table_path)
else:
X_in_table = aid.compute_features(candidates_in_table, verbose=1)
aid.save_features(X_in_table, features_in_table_path)
if os.path.exists(features_lookup_path):
X_lookup = aid.load_features(features_lookup_path)
else:
X_lookup = aid.compute_features(candidates_lookup, verbose=1)
aid.save_features(X_lookup, features_lookup_path)
#
if os.path.exists(final_candidates_ts_fields_path):
final_candidates_ts_fields = json.load(
open(final_candidates_ts_fields_path, 'rt'))
else:
predicted_ts_fields = [1 for c in candidates_ts_fields]
final_candidates_ts_fields = [
c for p, c in zip(predicted_ts_fields, candidates_ts_fields)
if p == 1
]
json.dump(final_candidates_ts_fields,
open(final_candidates_ts_fields_path, 'wt'),
indent=True)
if os.path.exists(final_candidates_in_table_path):
final_candidates_in_table = json.load(
open(final_candidates_in_table_path, 'rt'))
else:
predicted_in_table = aid.predict(X_in_table,
candidate_type=ev.CT_IN_TABLE)
final_candidates_in_table = [
c for p, c in zip(predicted_in_table, candidates_in_table)
if p == 1
]
json.dump(final_candidates_in_table,
open(final_candidates_in_table_path, 'wt'),
indent=True)
if os.path.exists(final_candidates_lookup_path):
final_candidates_lookup = json.load(
open(final_candidates_lookup_path, 'rt'))
else:
predicted_lookup = aid.predict(X_lookup, candidate_type=ev.CT_LOOKUP)
final_candidates_lookup = [
c for p, c in zip(predicted_lookup, candidates_lookup) if p == 1
]
json.dump(final_candidates_lookup,
open(final_candidates_lookup_path, 'wt'),
indent=True)
if build_events:
shutil.copyfile(mm, new_mm)
mm_engine_modif = ex.create_mm_engine(new_mm)
mm_meta_modif = ex.get_mm_meta(mm_engine_modif)
ev.compute_events(mm_engine_modif, mm_meta_modif,
final_candidates_ts_fields)
ev.compute_events(mm_engine_modif, mm_meta_modif,
final_candidates_in_table)
ev.compute_events(mm_engine_modif, mm_meta_modif,
final_candidates_lookup)
def info_log(mm_path, log_id):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
info = cn.log_info(mm_engine, mm_meta, log_id)
pprint(info)
def list_logs(mm_path):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
logs = cn.list_logs(mm_engine, mm_meta)
pprint(logs)
def case_notion_candidates_cached(mm_path,
dump_dir,
build_logs=False,
topk=None):
mm_engine = ex.create_mm_engine(mm_path)
mm_meta = ex.get_mm_meta(mm_engine)
os.makedirs(dump_dir, exist_ok=True)
class_stats_path = Path(dump_dir, 'class_stats_mm.json')
candidates_path = Path(dump_dir, 'candidates.pkl')
bounds_path = Path(dump_dir, 'bounds.json')
prediction_path = Path(dump_dir, 'prediction.json')
params_path = Path(dump_dir, 'params.json')
ranking_path = Path(dump_dir, 'ranking.json')
mm_modif_path = Path(dump_dir, 'mm-modif-logs.slexmm')
if os.path.isfile(class_stats_path):
class_stats = json.load(
open('{}/class_stats_mm.json'.format(dump_dir), 'rt'))
else:
class_stats = cn.get_stats_mm(mm_engine)
json.dump(class_stats,
open('{}/class_stats_mm.json'.format(dump_dir), 'wt'),
indent=True)
if os.path.isfile(candidates_path):
candidates_mem = cn.load_candidates(candidates_path)
else:
candidates = cn.compute_candidates(mm_engine, cache_dir=dump_dir)
candidates_mem = cn.save_candidates(candidates, candidates_path)
if os.path.isfile(bounds_path):
bounds = json.load(open(bounds_path, 'rt'))
else:
bounds = cn.compute_bounds_of_candidates(candidates_mem, mm_engine,
mm_meta, class_stats)
json.dump(bounds, open(bounds_path, 'wt'), indent=True)
if os.path.isfile(prediction_path):
pred = json.load(open(prediction_path, 'rt'))
else:
pred = cn.compute_prediction_from_bounds(bounds, 0.5, 0.5, 0.5)
json.dump(pred, open(prediction_path, 'wt'), indent=True)
if os.path.isfile(params_path):
params = json.load(open(params_path, 'rt'))
else:
params = {
'mode_sp': None,
'max_sp': None,
'min_sp': None,
'mode_lod': None,
'max_lod': 10,
'min_lod': 3,
'mode_ae': None,
'max_ae': 3000,
'min_ae': 0,
'w_sp': 0.33,
'w_lod': 0.33,
'w_ae': 0.33,
}
json.dump(params, open(params_path, 'wt'), indent=True)
pprint(params)
if os.path.isfile(ranking_path):
detailed_ranking = pd.DataFrame.from_csv(open(ranking_path, 'rt'))
else:
detailed_ranking = cn.compute_detailed_ranking(pred, **params)
detailed_ranking.to_csv(open(ranking_path, 'wt'))
ranking = detailed_ranking['cn_id']
pprint(detailed_ranking)
if build_logs:
shutil.copyfile(mm_path, mm_modif_path)
mm_engine_modif = ex.create_mm_engine(mm_modif_path)
mm_meta_modif = ex.get_mm_meta(mm_engine_modif)
if topk:
ranking_to_build = ranking[:topk]
else:
ranking_to_build = ranking
for idx in ranking_to_build:
c = candidates_mem[idx]
proc_name = 'proc_{}'.format(idx)
log_name = 'log_{}'.format(idx)
print('Building Log: {}'.format(log_name))
cn.build_log_for_case_notion(mm_engine_modif,
c,
proc_name=proc_name,
log_name=log_name,
metadata=mm_meta_modif)
def test_candidates():
shutil.copyfile(openslex_file_path_orig, openslex_file_path)
mm_engine = ex.create_mm_engine(openslex_file_path)
metadata = ex.get_mm_meta(mm_engine)
rs = cn.get_relationships(mm_engine)
assert rs.__len__() == 8
for r in rs:
assert r['rs'] in [
'CONCERT_HALL_FK', 'SEAT_HALL_FK', 'BP_CONCERT_FK', 'BP_BAND_FK',
'TICKET_CONCERT', 'TICKET_SEAT', 'BOOKING_CON', 'BOOKING_FK'
]
class_stats = cn.get_stats_mm(mm_engine)
os.makedirs('output/dumps/', exist_ok=True)
json.dump(class_stats,
open('output/dumps/stats_mm.json', 'wt'),
indent=True)
candidates = cn.compute_candidates(mm_engine)
json.dump(candidates,
open('output/dumps/candidates.json', 'wt'),
indent=True)
candidates = json.load(open('output/dumps/candidates.json', 'rt'),
object_hook=cn.CaseNotion)
for idx, cand in enumerate(candidates):
log_name = 'log_test_{}'.format(idx)
print('Computing Log: {}'.format(log_name))
log_id = cn.build_log_for_case_notion(
mm_engine,
cand,
proc_name='proc_test_{}'.format(idx),
log_name=log_name,
metadata=metadata)
sp = cn.compute_support_log(mm_engine, log_id, metadata)
print('Support for {}: {}'.format(log_id, sp))
ae = cn.compute_ae_log(mm_engine, log_id, metadata, sp)
print('AE for {}: {}'.format(log_id, ae))
lod = cn.compute_lod_log(mm_engine, log_id, metadata, sp)
print('LoD for {}: {}'.format(log_id, lod))
sp_lb = cn.compute_lb_support_cn(mm_engine, cand, metadata,
class_stats)
print('Lower Bound of Support for {}: {}'.format(idx, sp_lb))
sp_ub = cn.compute_ub_support_cn(mm_engine, cand, metadata,
class_stats)
print('Upper Bound of Support for {}: {}'.format(idx, sp_ub))
lod_lb = cn.compute_lb_lod_cn(mm_engine, cand, metadata, class_stats)
print('Lower Bound of LoD for {}: {}'.format(idx, lod_lb))
lod_ub = cn.compute_ub_lod_cn(mm_engine, cand, metadata, class_stats)
print('Upper Bound of LoD for {}: {}'.format(idx, lod_ub))
ae_lb = cn.compute_lb_ae_cn(mm_engine, cand, metadata, class_stats)
print('Lower Bound of AE for {}: {}'.format(idx, ae_lb))
ae_ub = cn.compute_ub_ae_cn(mm_engine, cand, metadata, class_stats)
print('Upper Bound of AE for {}: {}'.format(idx, ae_ub))
assert sp_ub >= sp >= sp_lb
assert lod_ub >= lod >= lod_lb
assert ae_ub >= ae >= ae_lb
assert candidates.__len__() > 0
def evaluate_cn(mm_filepath: str, output_dir):
os.makedirs(output_dir, exist_ok=True)
mm_filepath_tmp = '{}/mm-modif.slexmm'.format(output_dir)
df_filepath = '{}/df.json'.format(output_dir)
df_pickle_file = '{}/df.pickle'.format(output_dir)
class_stats_filepath = '{}/class_stats.json'.format(output_dir)
cands_filepath = '{}/candidates.json'.format(output_dir)
shutil.copyfile(mm_filepath, mm_filepath_tmp)
mm_engine = edex.create_mm_engine(mm_filepath_tmp)
mm_meta = edex.get_mm_meta(mm_engine)
cands = edcn.compute_candidates(mm_engine=mm_engine, max_length_path=5, cache_dir=output_dir)
stats = {
'proc_name': [],
'log_name': [],
'log_id': [],
'log_sp': [],
'log_lod': [],
'log_ae': [],
'cn_sp_lb': [],
'cn_sp_ub': [],
'cn_lod_lb': [],
'cn_lod_ub': [],
'cn_ae_lb': [],
'cn_ae_ub': [],
'num_classes': [],
'e': [],
'ir': [],
}
class_stats = edcn.get_stats_mm(mm_engine, mm_meta)
json.dump(class_stats, open(class_stats_filepath, mode='wt'), indent=True)
cn: CaseNotion
for i, cn in enumerate(tqdm(cands.values(), total=len(cands), desc='Candidates')):
proc_name = 'proc-{}'.format(i)
log_name = 'log-{}'.format(i)
log_id = edcn.build_log_for_case_notion(
mm_engine, cn, proc_name, log_name, mm_meta)
log_sp = edcn.compute_support_log(mm_engine, log_id, mm_meta)
log_lod = edcn.compute_lod_log(mm_engine, log_id, mm_meta)
log_ae = edcn.compute_ae_log(mm_engine, log_id, mm_meta)
cn_sp_lb = edcn.compute_lb_support_cn(mm_engine, cn, mm_meta, class_stats)
cn_sp_ub = edcn.compute_ub_support_cn(mm_engine, cn, mm_meta, class_stats)
cn_lod_lb = edcn.compute_lb_lod_cn(mm_engine, cn, mm_meta, class_stats)
cn_lod_ub = edcn.compute_ub_lod_cn(mm_engine, cn, mm_meta, class_stats)
cn_ae_lb = edcn.compute_lb_ae_cn(mm_engine, cn, mm_meta, class_stats)
cn_ae_ub = edcn.compute_ub_ae_cn(mm_engine, cn, mm_meta, class_stats)
stats['proc_name'].append(proc_name)
stats['log_name'].append(log_name)
stats['log_id'].append(log_id)
stats['log_sp'].append(log_sp)
stats['log_lod'].append(log_lod)
stats['log_ae'].append(log_ae)
stats['cn_sp_lb'].append(cn_sp_lb)
stats['cn_sp_ub'].append(cn_sp_ub)
stats['cn_lod_lb'].append(cn_lod_lb)
stats['cn_lod_ub'].append(cn_lod_ub)
stats['cn_ae_lb'].append(cn_ae_lb)
stats['cn_ae_ub'].append(cn_ae_ub)
stats['num_classes'].append(cn.get_classes_ids().__len__())
num_e = 0
sum_e_per_o = 0
for c_id in cn.get_classes_ids():
num_e = num_e + class_stats[str(c_id)]['e']
e_per_o = class_stats[str(c_id)]['ev_o']
num_o = class_stats[str(c_id)]['o_w_ev']
sum_e_per_o = float(sum_e_per_o) + float(float(e_per_o) / float(max(num_o, 1)))
stats['e'].append(num_e)
stats['ir'].append(sum_e_per_o / cn.get_classes_ids().__len__())
df = pd.DataFrame(stats)
df.to_json(df_filepath)
pickle.dump(df, open(df_pickle_file, mode='wb'))
def test_candidates_cached():
mm_engine_train = ex.create_mm_engine(train_openslex_file_path)
mm_meta_train = ex.get_mm_meta(mm_engine_train)
cached_dir_train = 'output/adw/ev_disc'
ts_train_path = '{}/timestamps.json'.format(cached_dir_train)
candidates_ts_fields_path = '{}/candidates_ts_fields.json'.format(
cached_dir_train)
candidates_in_table_path = '{}/candidates_in_table.json'.format(
cached_dir_train)
candidates_lookup_path = '{}/candidates_lookup.json'.format(
cached_dir_train)
features_in_table_path = '{}/features_in_table.json'.format(
cached_dir_train)
features_lookup_path = '{}/features_lookup.json'.format(cached_dir_train)
os.makedirs(cached_dir_train, exist_ok=True)
aid = ev.ActivityIdentifierDiscoverer(engine=mm_engine_train,
meta=mm_meta_train,
model='default')
if os.path.exists(ts_train_path):
timestamp_attrs = aid.load_timestamp_attributes(ts_train_path)
else:
timestamp_attrs = aid.get_timestamp_attributes()
aid.save_timestamp_attributes(timestamp_attrs, ts_train_path)
#
if os.path.exists(candidates_ts_fields_path):
candidates_ts_fields = aid.load_candidates(candidates_ts_fields_path)
else:
candidates_ts_fields = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_TS_FIELD)
aid.save_candidates(candidates_ts_fields, candidates_ts_fields_path)
if os.path.exists(candidates_in_table_path):
candidates_in_table = aid.load_candidates(candidates_in_table_path)
else:
candidates_in_table = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_IN_TABLE)
aid.save_candidates(candidates_in_table, candidates_in_table_path)
if os.path.exists(candidates_lookup_path):
candidates_lookup = aid.load_candidates(candidates_lookup_path)
else:
candidates_lookup = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_LOOKUP)
aid.save_candidates(candidates_lookup, candidates_lookup_path)
#
if os.path.exists(features_in_table_path):
X_in_table = aid.load_features(features_in_table_path)
else:
X_in_table = aid.compute_features(candidates_in_table, verbose=1)
aid.save_features(X_in_table, features_in_table_path)
if os.path.exists(features_lookup_path):
X_lookup = aid.load_features(features_lookup_path)
else:
X_lookup = aid.compute_features(candidates_lookup, verbose=1)
aid.save_features(X_lookup, features_lookup_path)
#
predicted_ts_fields = [1 for c in candidates_ts_fields]
predicted_in_table = aid.predict(X_in_table, candidate_type=ev.CT_IN_TABLE)
predicted_lookup = aid.predict(X_lookup, candidate_type=ev.CT_LOOKUP)
shutil.copyfile(train_openslex_file_path, modified_mm_path)
mm_engine_modif = ex.create_mm_engine(modified_mm_path)
mm_meta_modif = ex.get_mm_meta(mm_engine_modif)
pred_ts_cand = [
c for p, c in zip(predicted_ts_fields, candidates_ts_fields) if p == 1
]
pred_in_table_cand = [
c for p, c in zip(predicted_in_table, candidates_in_table) if p == 1
]
pred_lookup_cand = [
c for p, c in zip(predicted_lookup, candidates_lookup) if p == 1
]
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_ts_cand[:2])
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_in_table_cand[:2])
ev.compute_events(mm_engine_modif, mm_meta_modif, pred_lookup_cand[:2])
def test_trained_model_cached(openslex_train=train_openslex_file_path,
ground_truth_train=ground_truth_path,
openslex_test=train_openslex_file_path,
ground_truth_test=ground_truth_path,
cached_dir_train='output/A/ev_disc',
cached_dir_test='output/B/ev_disc'):
ts_train_path = '{}/timestamps.json'.format(cached_dir_train)
candidates_ts_fields_path = '{}/candidates_ts_fields.json'.format(
cached_dir_train)
candidates_in_table_path = '{}/candidates_in_table.json'.format(
cached_dir_train)
candidates_lookup_path = '{}/candidates_lookup.json'.format(
cached_dir_train)
features_in_table_path = '{}/features_in_table.json'.format(
cached_dir_train)
features_lookup_path = '{}/features_lookup.json'.format(cached_dir_train)
model_trained_path = '{}/model.pkl'.format(cached_dir_train)
ts_test_path = '{}/timestamps.json'.format(cached_dir_test)
candidates_test_ts_fields_path = '{}/candidates_ts_fields.json'.format(
cached_dir_test)
candidates_test_in_table_path = '{}/candidates_in_table.json'.format(
cached_dir_test)
candidates_test_lookup_path = '{}/candidates_lookup.json'.format(
cached_dir_test)
features_test_in_table_path = '{}/features_in_table.json'.format(
cached_dir_test)
features_test_lookup_path = '{}/features_lookup.json'.format(
cached_dir_test)
if not os.path.exists(cached_dir_test):
os.makedirs(cached_dir_test)
if not os.path.exists(cached_dir_train):
os.makedirs(cached_dir_train)
mm_engine_train = ex.create_mm_engine(openslex_train)
mm_meta_train = ex.get_mm_meta(mm_engine_train)
aid = ev.ActivityIdentifierDiscoverer(engine=mm_engine_train,
meta=mm_meta_train,
model=None)
if os.path.exists(ts_train_path):
timestamp_attrs = aid.load_timestamp_attributes(ts_train_path)
else:
timestamp_attrs = aid.get_timestamp_attributes()
aid.save_timestamp_attributes(timestamp_attrs, ts_train_path)
#
if os.path.exists(candidates_ts_fields_path):
candidates_ts_fields = aid.load_candidates(candidates_ts_fields_path)
else:
candidates_ts_fields = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_TS_FIELD)
aid.save_candidates(candidates_ts_fields, candidates_ts_fields_path)
if os.path.exists(candidates_in_table_path):
candidates_in_table = aid.load_candidates(candidates_in_table_path)
else:
candidates_in_table = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_IN_TABLE)
aid.save_candidates(candidates_in_table, candidates_in_table_path)
if os.path.exists(candidates_lookup_path):
candidates_lookup = aid.load_candidates(candidates_lookup_path)
else:
candidates_lookup = aid.generate_candidates(
timestamp_attrs=timestamp_attrs, candidate_type=ev.CT_LOOKUP)
aid.save_candidates(candidates_lookup, candidates_lookup_path)
#
if os.path.exists(features_in_table_path):
X_in_table = aid.load_features(features_in_table_path)
else:
X_in_table = aid.compute_features(candidates_in_table, verbose=1)
aid.save_features(X_in_table, features_in_table_path)
if os.path.exists(features_lookup_path):
X_lookup = aid.load_features(features_lookup_path)
else:
X_lookup = aid.compute_features(candidates_lookup, verbose=1)
aid.save_features(X_lookup, features_lookup_path)
#
y_true_train_in_table = aid.load_y_true(candidates_in_table,
y_true_path=ground_truth_train)
y_true_train_lookup = aid.load_y_true(candidates_lookup,
y_true_path=ground_truth_train)
try:
class_weight_in_table = compute_class_weight('balanced', [0, 1],
y_true_train_in_table)
except:
class_weight_in_table = [1.0, 1.0]
try:
class_weight_lookup = compute_class_weight('balanced', [0, 1],
y_true_train_lookup)
except:
class_weight_lookup = [1.0, 1.0]
classifiers = {
ev.CT_IN_TABLE:
ev.make_sklearn_pipeline(
XGBClassifier(max_depth=2,
n_estimators=10,
random_state=1,
scale_pos_weight=class_weight_in_table[1])),
ev.CT_LOOKUP:
ev.make_sklearn_pipeline(
XGBClassifier(max_depth=2,
n_estimators=10,
random_state=1,
scale_pos_weight=class_weight_lookup[1]))
}
aid.set_model(classifiers)
aid.train_model(X_in_table,
y_true_train_in_table,
candidate_type=ev.CT_IN_TABLE)
aid.train_model(X_lookup, y_true_train_lookup, candidate_type=ev.CT_LOOKUP)
y_pred_ts_fields = [1 for c in candidates_ts_fields]
y_pred_in_table = aid.predict(X_in_table, candidate_type=ev.CT_IN_TABLE)
y_pred_lookup = aid.predict(X_lookup, candidate_type=ev.CT_LOOKUP)
scores_train_in_table = aid.score(y_true_train_in_table, y_pred_in_table)
scores_train_lookup = aid.score(y_true_train_lookup, y_pred_lookup)
print('Scores In Table')
pprint(scores_train_in_table)
print('Scores Lookup')
pprint(scores_train_lookup)
model_trained = classifiers
with open(model_trained_path, mode='wb') as f:
pickle.dump(classifiers, f)
mm_engine_test = ex.create_mm_engine(openslex_test)
mm_meta_test = ex.get_mm_meta(mm_engine_test)
aid_test = ev.ActivityIdentifierDiscoverer(engine=mm_engine_test,
meta=mm_meta_test,
model=model_trained)
if os.path.exists(ts_test_path):
timestamp_attrs_test = aid_test.load_timestamp_attributes(ts_test_path)
else:
timestamp_attrs_test = aid_test.get_timestamp_attributes()
aid_test.save_timestamp_attributes(timestamp_attrs_test, ts_test_path)
#
if os.path.exists(candidates_test_ts_fields_path):
candidates_test_ts_fields = aid_test.load_candidates(
candidates_test_ts_fields_path)
else:
candidates_test_ts_fields = aid_test.generate_candidates(
timestamp_attrs=timestamp_attrs_test,
candidate_type=ev.CT_IN_TABLE)
aid_test.save_candidates(candidates_test_ts_fields,
candidates_test_ts_fields_path)
if os.path.exists(candidates_test_in_table_path):
candidates_test_in_table = aid_test.load_candidates(
candidates_test_in_table_path)
else:
candidates_test_in_table = aid_test.generate_candidates(
timestamp_attrs=timestamp_attrs_test,
candidate_type=ev.CT_IN_TABLE)
aid_test.save_candidates(candidates_test_in_table,
candidates_test_in_table_path)
if os.path.exists(candidates_test_lookup_path):
candidates_test_lookup = aid_test.load_candidates(
candidates_test_lookup_path)
else:
candidates_test_lookup = aid_test.generate_candidates(
timestamp_attrs=timestamp_attrs_test, candidate_type=ev.CT_LOOKUP)
aid_test.save_candidates(candidates_test_lookup,
candidates_test_lookup_path)
#
if os.path.exists(features_test_in_table_path):
feature_values_in_table_test = aid_test.load_features(
features_test_in_table_path)
else:
feature_values_in_table_test = aid_test.compute_features(
candidates_test_in_table, verbose=True)
aid_test.save_features(feature_values_in_table_test,
features_test_in_table_path)
if os.path.exists(features_test_lookup_path):
feature_values_lookup_test = aid_test.load_features(
features_test_lookup_path)
else:
feature_values_lookup_test = aid_test.compute_features(
candidates_test_lookup, verbose=True)
aid_test.save_features(feature_values_lookup_test,
features_test_lookup_path)
#
pred_test_ts_fields = [1 for c in candidates_test_ts_fields]
pred_test_in_table = aid_test.predict(feature_values_in_table_test,
candidate_type=ev.CT_IN_TABLE)
pred_test_lookup = aid_test.predict(feature_values_lookup_test,
candidate_type=ev.CT_LOOKUP)
y_true_in_table = aid_test.load_y_true(candidates_test_in_table,
ground_truth_test)
y_true_lookup = aid_test.load_y_true(candidates_test_lookup,
ground_truth_test)
y_true_ts_fields = aid_test.load_y_true(candidates_test_ts_fields,
ground_truth_test)
scores_ts_fields = aid_test.score(y_true_ts_fields, pred_test_ts_fields)
scores_in_table = aid_test.score(y_true_in_table, pred_test_in_table)
scores_lookup = aid_test.score(y_true_lookup, pred_test_lookup)
print('Score Ts Fields')
pprint(scores_ts_fields)
print('Score In Table')
pprint(scores_in_table)
print('Score Lookup')
pprint(scores_lookup)