def count_unique_people(tree_files):
all_xrefs = set()
for tf in tree_files:
_, people_dict = person.read_people(tf, clean=True)
all_xrefs |= set(people_dict.keys())
print "{} unique xrefs.".format(len(all_xrefs))
return len(all_xrefs)
def count_unique_people(tree_files):
all_xrefs = set()
for tf in tree_files:
_, people_dict = person.read_people(tf, clean=True)
all_xrefs |= set(people_dict.keys())
print "{} unique xrefs.".format(len(all_xrefs))
return len(all_xrefs)
def experiment_scalability(n_reps=1, n_trees=5, n_people=500,
varied_param='n_people', params=[50,100,500],
methods=('unary', 'LD', 'mKlau'), top_k_matches=5,
f=1.0, title='scalability', do_save=True,
rep_offset=0, subfolder=''):
nvv = len(params)
res_precision = np.zeros((len(methods), nvv, n_reps))
res_recall = np.zeros((len(methods), nvv, n_reps))
res_fscore = np.zeros((len(methods), nvv, n_reps))
res_t = np.zeros((len(methods), nvv, n_reps))
res_iterations = np.zeros((len(methods), nvv, n_reps))
res_clusters = np.zeros((len(methods), nvv, n_reps))
res_lb = np.zeros((len(methods), nvv, n_reps))
res_ub = np.zeros((len(methods), nvv, n_reps))
t_beg = time.time()
start_date_part = str(dt.datetime.now())[:19]
start_date_part = re.sub(' ', '_', start_date_part)
start_date_part = re.sub(':', '', start_date_part)
dir_path = os.path.join("experiment_results", subfolder)
if not os.path.exists(dir_path):
os.makedirs(dir_path)
fname0 = os.path.join(dir_path, "{}_part_{}.pckl".format(
title, start_date_part))
for r in range(n_reps):
print "\n--- Repetition {}. ---".format(r+1)
for i, param_val in enumerate(params):
if varied_param == 'n_people':
n_people = param_val
elif varied_param == 'n_trees':
n_trees = param_val
else:
raise Exception('Invalid varied parameter: ' + varied_param)
# Generate data
try:
tree_files = extract_ft.get_k_fragments(
n_trees, n_people, label="first{}".format(r+rep_offset))
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
except:
# If it fails, try generating new trees.
tree_files = extract_ft.get_k_fragments(
n_trees, n_people, label="first{}".format(r+rep_offset),
check_if_exists=False)
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
print "\n rep={}, {}={}".format(r+1, varied_param, param_val)
for mi, m in enumerate(methods):
print "\n rep={}, {}={}, method={}".format(r+1, varied_param,
param_val, m)
t0 = time.time()
precision, recall, fscore, n_clusters, lb, ub, iters = \
merge_multiple(people_index_tuples, 10, top_k_matches,
method=m, uniq_people=uniq_people, f=f)
res_precision[mi, i, r] = precision
res_recall[mi, i, r] = recall
res_fscore[mi, i, r] = fscore
res_clusters[mi, i, r] = n_clusters
res_t[mi, i, r] = time.time() - t0
res_iterations[mi, i, r] = iters
res_lb[mi, i, r] = lb
res_ub[mi, i, r] = ub
if do_save and n_reps > 1:
pickle.dump(locals(), open(fname0, 'wb'))
print "Wrote the results of repetition {} to: {}\n".format(r+1,
fname0)
print "\nThe whole experiment took {:2f} seconds.".format(time.time()-t_beg)
date_part = str(dt.datetime.now())[:19]
date_part = re.sub(' ', '_', date_part)
date_part = re.sub(':', '', date_part)
fname = os.path.join(dir_path, "{}_{}.pckl".format(
title, date_part))
if do_save:
pickle.dump(locals(), open(fname, 'wb'))
print "Wrote the results to: {}\n".format(fname)
print "F1 score:", np.mean(res_fscore, axis=2)
print "Precision:", np.mean(res_precision, axis=2)
print "Recall:", np.mean(res_recall, axis=2)
print "Time:", np.mean(res_t, axis=2)
print "Clusters:", np.mean(res_clusters, axis=2)
print "Lower bounds:", np.mean(res_lb, axis=2)
print "Upper bounds:", np.mean(res_ub, axis=2)
def experiment_multiple_trees(n_reps=1, n_trees=5, n_people=500,
methods=('unary', 'LD', 'mKlau'),
top_k_matches=5, f_vals=(0.1, 0.5, 1, 1.5, 2),
title='genealogical', do_save=True, dir_id=None,
rep_offset=0):
nvv = len(f_vals)
res_precision = np.zeros((len(methods), nvv, n_reps))
res_recall = np.zeros((len(methods), nvv, n_reps))
res_fscore = np.zeros((len(methods), nvv, n_reps))
res_t = np.zeros((len(methods), nvv, n_reps))
res_iterations = np.zeros((len(methods), nvv, n_reps))
res_clusters = np.zeros((len(methods), nvv, n_reps))
res_lb = np.zeros((len(methods), nvv, n_reps))
res_ub = np.zeros((len(methods), nvv, n_reps))
t_beg = time.time()
start_date_part = str(dt.datetime.now())[:19]
start_date_part = re.sub(' ', '_', start_date_part)
start_date_part = re.sub(':', '', start_date_part)
fname0 = os.path.join("experiment_results", "{}_part_{}.pckl".format(
title, start_date_part))
for r in range(n_reps):
print "\n--- Repetition {}. ---".format(r+1)
# Generate data
tree_files = extract_ft.get_k_fragments(
n_trees, n_people, label="first{}".format(r+rep_offset))
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
for i, f in enumerate(f_vals):
print "\n rep={}, f={}".format(r+1, f)
for mi, m in enumerate(methods):
if m.startswith('meLD') and i > 0:
# No need to compute fixed entity method for different f values.
continue
print "\n rep={}, f={}, method={}\n".format(r+1, f, m)
t0 = time.time()
precision, recall, fscore, n_clusters, lb, ub, iters = \
merge_multiple(people_index_tuples, 10, top_k_matches,
method=m, uniq_people=uniq_people, f=f)
res_precision[mi, i, r] = precision
res_recall[mi, i, r] = recall
res_fscore[mi, i, r] = fscore
res_clusters[mi, i, r] = n_clusters
res_t[mi, i, r] = time.time() - t0
res_iterations[mi, i, r] = iters
res_lb[mi, i, r] = lb
res_ub[mi, i, r] = ub
if do_save and n_reps > 1:
pickle.dump(locals(), open(fname0, 'wb'))
print "Wrote the results of repetition {} to: {}\n".format(r+1, fname0)
print "\nThe whole experiment took {:2f} seconds.".format(time.time()-t_beg)
if do_save:
fname = util.save_data(locals(), title, dir_name='genealogy{}'.format(
str(dir_id)))
print "Wrote the results to: {}".format(fname)
print "F1 score:", np.mean(res_fscore, axis=2)
print "Precision:", np.mean(res_precision, axis=2)
print "Recall:", np.mean(res_recall, axis=2)
print "Time:", np.mean(res_t, axis=2)
print "Clusters:", np.mean(res_clusters, axis=2)
print "Lower bounds:", np.mean(res_lb, axis=2)
print "Upper bounds:", np.mean(res_ub, axis=2)
def experiment_multiple_trees(n_reps=1,
n_trees=5,
n_people=500,
methods=('unary', 'LD', 'mKlau'),
top_k_matches=5,
f_vals=(0.1, 0.5, 1, 1.5, 2)):
nvv = len(f_vals)
res_precision = np.zeros((len(methods), nvv, n_reps))
res_recall = np.zeros((len(methods), nvv, n_reps))
res_fscore = np.zeros((len(methods), nvv, n_reps))
res_t = np.zeros((len(methods), nvv, n_reps))
res_iterations = np.zeros((len(methods), nvv, n_reps))
res_clusters = np.zeros((len(methods), nvv, n_reps))
res_lb = np.zeros((len(methods), nvv, n_reps))
res_ub = np.zeros((len(methods), nvv, n_reps))
t_beg = time.time()
start_date_part = str(dt.datetime.now())[:19]
start_date_part = re.sub(' ', '_', start_date_part)
start_date_part = re.sub(':', '', start_date_part)
fname0 = os.path.join("experiment_results",
"genealogical_{}.pckl".format(start_date_part))
for j in range(n_reps):
print "\n--- Repetition {}. ---".format(j + 1)
# Generate data
tree_files = extract_ft.get_k_fragments(n_trees,
n_people,
label="first{}".format(j))
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
for i, f in enumerate(f_vals):
print "\n rep={}, f={}".format(j + 1, f)
for mi, m in enumerate(methods):
print "\n rep={}, f={}, method={}\n".format(j + 1, f, m)
t0 = time.time()
precision, recall, fscore, n_clusters, lb, ub, iters = \
merge_multiple(people_index_tuples, 10, top_k_matches,
method=m, uniq_people=uniq_people, f=f)
res_precision[mi, i, j] = precision
res_recall[mi, i, j] = recall
res_fscore[mi, i, j] = fscore
res_clusters[mi, i, j] = n_clusters
res_t[mi, i, j] = time.time() - t0
res_iterations[mi, i, j] = iters
res_lb[mi, i, j] = lb
res_ub[mi, i, j] = ub
pickle.dump(locals(), open(fname0, 'wb'))
print "Wrote the results of repetition {} to: {}\n".format(
j + 1, fname0)
print "\nThe whole experiment took {:2f} seconds.".format(time.time() -
t_beg)
date_part = str(dt.datetime.now())[:19]
date_part = re.sub(' ', '_', date_part)
date_part = re.sub(':', '', date_part)
fname = os.path.join("experiment_results",
"genealogical_{}.pckl".format(date_part))
pickle.dump(locals(), open(fname, 'wb'))
print "Wrote the results to: {}\n".format(fname)
print "F1 score:", np.mean(res_fscore, axis=2)
print "Precision:", np.mean(res_precision, axis=2)
print "Recall:", np.mean(res_recall, axis=2)
print "Time:", np.mean(res_t, axis=2)
print "Clusters:", np.mean(res_clusters, axis=2)
print "Lower bounds:", np.mean(res_lb, axis=2)
print "Upper bounds:", np.mean(res_ub, axis=2)
def experiment_scalability(n_reps=1,
n_trees=5,
n_people=500,
varied_param='n_people',
params=[50, 100, 500],
methods=('unary', 'LD', 'mKlau'),
top_k_matches=5,
f=1.0,
title='scalability',
do_save=True,
rep_offset=0,
subfolder=''):
nvv = len(params)
res_precision = np.zeros((len(methods), nvv, n_reps))
res_recall = np.zeros((len(methods), nvv, n_reps))
res_fscore = np.zeros((len(methods), nvv, n_reps))
res_t = np.zeros((len(methods), nvv, n_reps))
res_iterations = np.zeros((len(methods), nvv, n_reps))
res_clusters = np.zeros((len(methods), nvv, n_reps))
res_lb = np.zeros((len(methods), nvv, n_reps))
res_ub = np.zeros((len(methods), nvv, n_reps))
t_beg = time.time()
start_date_part = str(dt.datetime.now())[:19]
start_date_part = re.sub(' ', '_', start_date_part)
start_date_part = re.sub(':', '', start_date_part)
dir_path = os.path.join("experiment_results", subfolder)
if not os.path.exists(dir_path):
os.makedirs(dir_path)
fname0 = os.path.join(dir_path,
"{}_part_{}.pckl".format(title, start_date_part))
for r in range(n_reps):
print "\n--- Repetition {}. ---".format(r + 1)
for i, param_val in enumerate(params):
if varied_param == 'n_people':
n_people = param_val
elif varied_param == 'n_trees':
n_trees = param_val
else:
raise Exception('Invalid varied parameter: ' + varied_param)
# Generate data
try:
tree_files = extract_ft.get_k_fragments(
n_trees, n_people, label="first{}".format(r + rep_offset))
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
except:
# If it fails, try generating new trees.
tree_files = extract_ft.get_k_fragments(
n_trees,
n_people,
label="first{}".format(r + rep_offset),
check_if_exists=False)
people_index_tuples = []
for tf in tree_files:
people, people_dict = person.read_people(tf, clean=True)
#'family_trees/data/rand_frag_%d/' % i, clean=True)
index = create_index(people)
people_index_tuples.append((people, index, people_dict))
uniq_people = count_unique_people(tree_files)
print "\n rep={}, {}={}".format(r + 1, varied_param, param_val)
for mi, m in enumerate(methods):
print "\n rep={}, {}={}, method={}".format(
r + 1, varied_param, param_val, m)
t0 = time.time()
precision, recall, fscore, n_clusters, lb, ub, iters = \
merge_multiple(people_index_tuples, 10, top_k_matches,
method=m, uniq_people=uniq_people, f=f)
res_precision[mi, i, r] = precision
res_recall[mi, i, r] = recall
res_fscore[mi, i, r] = fscore
res_clusters[mi, i, r] = n_clusters
res_t[mi, i, r] = time.time() - t0
res_iterations[mi, i, r] = iters
res_lb[mi, i, r] = lb
res_ub[mi, i, r] = ub
if do_save and n_reps > 1:
pickle.dump(locals(), open(fname0, 'wb'))
print "Wrote the results of repetition {} to: {}\n".format(
r + 1, fname0)
print "\nThe whole experiment took {:2f} seconds.".format(time.time() -
t_beg)
date_part = str(dt.datetime.now())[:19]
date_part = re.sub(' ', '_', date_part)
date_part = re.sub(':', '', date_part)
fname = os.path.join(dir_path, "{}_{}.pckl".format(title, date_part))
if do_save:
pickle.dump(locals(), open(fname, 'wb'))
print "Wrote the results to: {}\n".format(fname)
print "F1 score:", np.mean(res_fscore, axis=2)
print "Precision:", np.mean(res_precision, axis=2)
print "Recall:", np.mean(res_recall, axis=2)
print "Time:", np.mean(res_t, axis=2)
print "Clusters:", np.mean(res_clusters, axis=2)
print "Lower bounds:", np.mean(res_lb, axis=2)
print "Upper bounds:", np.mean(res_ub, axis=2)