def __init__(self, feature_template=None, actionxid_map=None):
self.feature_template = feature_template
self.actionxid_map = actionxid_map
if actionxid_map is not None:
self.idxaction_map = reverse_dict(actionxid_map)
# self.classifier = RandomForestClassifier(n_estimators=3, n_jobs=-1)
self.classifier = LinearSVC(C=1.0, penalty='l2', loss='squared_hinge', dual=False, tol=1e-7, max_iter=10000)
def __init__(self,
feature_template_level_0=None,
feature_template_level_1=None,
feature_template_level_2=None,
relationxid_map=None):
self.feature_template_level_0 = feature_template_level_0
self.feature_template_level_1 = feature_template_level_1
self.feature_template_level_2 = feature_template_level_2
self.relationxid_map = relationxid_map
if relationxid_map is not None:
self.idxrelation_map = reverse_dict(relationxid_map)
self.classifier_level_0 = LinearSVC(C=1.0,
penalty='l1',
loss='squared_hinge',
dual=False,
tol=1e-7)
self.classifier_level_1 = LinearSVC(C=1.0,
penalty='l1',
loss='squared_hinge',
dual=False,
tol=1e-7)
self.classifier_level_2 = LinearSVC(C=1.0,
penalty='l1',
loss='squared_hinge',
dual=False,
tol=1e-7)
def load(self, fname):
""" Load models
"""
with gzip.open(fname, 'rb') as fin:
data = pickle.load(fin)
self.classifier = data['action_clf']
self.feature_template = data['feature_template']
self.idxaction_map = data['idxaction_map']
self.actionxid_map = reverse_dict(self.idxaction_map)
print('Load action classifier from file: {} with {} features and {} actions.'.format(fname,
len(self.feature_template),
len(self.idxaction_map)))
def __init__(self, data_helper, config):
super(NeuralClassifier, self).__init__()
self.data_helper = data_helper
self.xidx_action_map = reverse_dict(data_helper.action_map)
self.config = config
self.init_embeddings()
self.tokenizer = RobertaTokenizer.from_pretrained('distilroberta-base')
self.bert = RobertaModel.from_pretrained('distilroberta-base')
if self.config[UNTRAINED_ROBERTA]:
# Initialize new model with the same config as distilroberta
self.bert = RobertaModel(self.bert.config)
self.out_classifier = nn.Sequential(
nn.Linear(self.get_classifier_dim(), self.config[HIDDEN_DIM]),
nn.GELU(),
nn.Linear(self.config[HIDDEN_DIM], 4)
)
def load(self, fname):
""" Load models
"""
with gzip.open(fname, 'rb') as fin:
data = pickle.load(fin)
self.classifier_level_0 = data['relation_clf_level_0']
self.classifier_level_1 = data['relation_clf_level_1']
self.classifier_level_2 = data['relation_clf_level_2']
self.feature_template_level_0 = data['feature_template_level_0']
self.feature_template_level_1 = data['feature_template_level_1']
self.feature_template_level_2 = data['feature_template_level_2']
self.idxrelation_map = data['idxrelation_map']
self.relationxid_map = reverse_dict(self.idxrelation_map)
print(
'Load relation classifier from file: {} with {} features at level 0, '
'{} features at level 1, {} features at level 2, and {} relations.'
.format(fname, len(self.feature_template_level_0),
len(self.feature_template_level_1),
len(self.feature_template_level_2),
len(self.idxrelation_map)))