def eval_and_log(evl):
print('Macro Precision: {}'.format(evl.macro_precision))
log_metric('Macro Precision', evl.macro_precision)
print('Macro Recall: {}'.format(evl.macro_recall))
log_metric('Macro Recall', evl.macro_recall)
print('Macro F1: {}'.format(evl.macro_f1_score))
log_metric('Macro F1', evl.macro_f1_score)
print('Micro Precision: {}'.format(evl.micro_precision))
log_metric('Micro Precision', evl.micro_precision)
print('Micro Recall: {}'.format(evl.micro_recall))
log_metric('Micro Recall', evl.micro_recall)
print('Micro F1: {}'.format(evl.micro_f1_score))
log_metric('Micro F1', evl.micro_f1_score)
print('Pairwise Precision: {}'.format(evl.pairwise_precision))
log_metric('Pairwise Precision', evl.pairwise_precision)
print('Pairwise Recall: {}'.format(evl.pairwise_recall))
print('Pairwise F1: {}'.format(evl.pairwise_f1_score))
log_metric('Pairwise F1', evl.pairwise_f1_score)
log_artifact(get_abspath(LOG_PATH))
def visualize_cluster(cfg):
file_name = cfg.ex.file_name
enc_name = cfg.ex.enc
enc_model = LMS[enc_name]()
np_n_layer = cfg.model.np_n_layer or enc_model.default_max_layer
rp_n_layer = cfg.model.rp_n_layer or enc_model.default_max_layer
print('--- Start: build model ---')
model = build_model(enc_model=enc_model, file_name=file_name)
print('--- End: build model ---')
print('--- Start: read phrases ---')
entities, relations = model.get_encoded_elems(np_n_layer, rp_n_layer)
print('--- End: read phrases ---')
print('--- Start: read clusters ---')
ent2clusters, rel2clusters = model.read_clusters()
print('--- End: read clusters ---')
plt_dir = f'plt_img/{model.cluster_dumped_dir}'
os.makedirs(get_abspath(plt_dir), exist_ok=True)
plt_path = f'{plt_dir}/{model.cluster_file_name}'
n_min_elems = cfg.vis.n_min_elems
n_max_elems = cfg.vis.n_max_elems
if n_min_elems:
plt_path += f'_min{n_min_elems}'
if n_max_elems:
plt_path += f'_max{n_max_elems}'
plt_path += f'.png'
fig, axes = plt.subplots(1, 2, figsize=(20, 10))
fig.suptitle('t-SNE of entities and relation clusters')
print('--- Start: plot noun phrases ---')
scatter_tsne(entities, ent2clusters, axes[0], n_max_elems, n_min_elems)
print('--- End: plot noun phrases ---')
print('--- Start: plot rel phrases ---')
scatter_tsne(relations, rel2clusters, axes[1], n_max_elems, n_min_elems)
print('--- End: plot rel phrases ---')
plt.savefig(get_abspath(plt_path))
def embed_elements_path(file_prefix, pretrained_name):
return get_abspath(f'{ELEM_FILE_PATH}/{file_prefix}_{pretrained_name}.pkl')
def read_clusters(self, cluster_dumped_path):
with open(hlp.get_abspath(cluster_dumped_path), 'rb') as f:
return pickle.load(f)
def dump_clusters(self, clusters):
os.makedirs(hlp.get_abspath(self.cluster_dumped_dir), exist_ok=True)
with open(hlp.get_abspath(self.cluster_dumped_path), 'wb') as f:
pickle.dump(clusters, f, protocol=pickle.HIGHEST_PROTOCOL)
def __init__(self, model, file_name):
self.model = model
self.file_name = file_name
file_path = hlp.get_abspath(f'{DATA_PATH}/{file_name}')
self.data = Triples.from_file(file_path)
self.__init_elements()