def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = [
'CR', 'MR', 'MPQA', 'SUBJ', 'SST2', 'SST5', 'TREC', 'MRPC',
'SICKRelatedness', 'SICKEntailment', 'STSBenchmark', 'SNLI',
'ImageCaptionRetrieval', 'STS12', 'STS13', 'STS14', 'STS15',
'STS16'
]
def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = [
'BioC', 'CitationSA', 'ClinicalSA', 'BioASQ', 'PICO', 'PubMed20K',
'RQE', 'ClinicalSTS', 'BIOSSES', 'MedNLI'
]
def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = [
'CR', 'MR', 'MPQA', 'SUBJ', 'SST2', 'SST5', 'TREC', 'MRPC',
'SICKRelatedness', 'SICKEntailment', 'STSBenchmark', 'SNLI',
'ImageCaptionRetrieval', 'STS12', 'STS13', 'STS14', 'STS15',
'STS16', 'Length', 'WordContent', 'Depth', 'TopConstituents',
'BigramShift', 'Tense', 'SubjNumber', 'ObjNumber', 'OddManOut',
'CoordinationInversion', 'AmenitySimilarEvents'
]
def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = ['STS12', 'STS13', 'STS14', 'STS15', 'STS16']
def apply_logician(s1, s2 , is_list=False, sick_model = False):
# is_list : If you are directly sending sentences then keep is_list = False
# If you are sending list of list of words then keep is_list = True
# sick_model: if True, will use sick model for prediction
# : if False, will use snli model for prediction
# Load InferSent model
params_model = {'bsize': 64, 'word_emb_dim': 300, 'enc_lstm_dim': 2048,
'pool_type': 'max', 'dpout_model': 0.0, 'version': V}
model = InferSent(params_model)
model.load_state_dict(torch.load(MODEL_PATH))
model.set_w2v_path(PATH_TO_W2V)
params_senteval = {'task_path': PATH_TO_DATA, 'usepytorch': True, 'kfold': 5}
params_senteval['classifier'] = {'nhid': 0, 'optim': 'rmsprop', 'batch_size': 128,
'tenacity': 3, 'epoch_size': 2}
params_senteval['infersent'] = model.cuda()
if not is_list:
s1 = convert_str2lst(s1)
s2 = convert_str2lst(s2)
samples = s1+s2
params_senteval['batch_size'] = min(128,len(s1))
params_senteval = utils.dotdict(params_senteval)
params_senteval.usepytorch = True
prepare(params_senteval, samples)
emb_s1 = batcher(params_senteval, s1)
emb_s2 = batcher(params_senteval, s2)
if sick_model:
testF = np.c_[ np.abs(emb_s1 - emb_s2),emb_s1 * emb_s2]
cp = torch.load('./saved_sick.pth')
print('[Contradiction Neutral Entailment]')
else:
testF = np.c_[emb_s1, emb_s2, emb_s1 * emb_s2, np.abs(emb_s1 - emb_s2)]
cp = torch.load('./saved_snli_augment_ordered.pth')
print('[ Entailment Neutral Contradiction ]')
inputdim = testF.shape[1]
nclasses = 3
clf = nn.Sequential(nn.Linear(inputdim, nclasses),).cuda()
clf.load_state_dict(cp)
testF = torch.FloatTensor(testF).cuda()
out = clf(testF)
sf = nn.Softmax(1)
probs = sf(out)
return probs
def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.usepytorch = True if "usepytorch" not in params else params.usepytorch
params.seed = 1111 if "seed" not in params else params.seed
params.batch_size = 128 if "batch_size" not in params else params.batch_size
params.nhid = 0 if "nhid" not in params else params.nhid
params.kfold = 5 if "kfold" not in params else params.kfold
if "classifier" not in params or not params["classifier"]:
params.classifier = {"nhid": 0}
if "nhid" not in params.classifier:
raise ValueError(
"Number of hidden units not set. Please set number of hidden units "
+ "in classifier config.")
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = [
# "Length",
# "EntityDistance",
# "ArgumentOrder",
# "EntityExistsBetweenHeadTail",
# "EntityCountORGBetweenHeadTail",
# "EntityCountPERBetweenHeadTail",
# "EntityCountDATEBetweenHeadTail",
# "EntityCountMISCBetweenHeadTail",
# "EntityCountLOCBetweenHeadTail",
# "PosTagHeadLeft",
# "PosTagHeadRight",
# "PosTagTailLeft",
# "PosTagTailRight",
# "ArgTypeHead",
# "ArgTypeTail",
# "TreeDepth",
# "SDPTreeDepth",
# "ArgumentHeadGrammaticalRole",
# "ArgumentTailGrammaticalRole",
# Kirk's new code
"ArgumentAddGrammarRole_Head",
"ArgumentAddGrammarRole_Tail",
"ArgumentGrammarRole_ControlHead",
"ArgumentGrammarRole_ControlTail",
]
def __init__(self, params, batcher, prepare=None):
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = [
'WCCRS_HOTELS', 'WCCRS_MEDICINE', 'SICKRelatedness',
'SICKEntailment', '8TAGS'
]
def __init__(self, params, batcher, prepare=None):
# parameters
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
# batcher and prepare
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
self.list_tasks = ['SST2', 'SST3', 'MRPC', 'ReadabilityCl', 'TagCl', 'PoemsCl', 'ProzaCl', 'TREC', 'STS', 'SICK']
def __init__(self, params, batcher, prepare=None):
# setting default parameters
params = utils.dotdict(params)
params.usepytorch = True if 'usepytorch' not in params else params.usepytorch
params.seed = 1111 if 'seed' not in params else params.seed
params.batch_size = 128 if 'batch_size' not in params else params.batch_size
params.nhid = 0 if 'nhid' not in params else params.nhid
params.kfold = 5 if 'kfold' not in params else params.kfold
if 'classifier' not in params or not params['classifier']:
params.classifier = {'nhid': 0}
assert 'nhid' in params.classifier, 'Set number of hidden units in classifier config!!'
self.params = params
# set up bilinear projection, with learnable matrix W
params.bilinear = False if 'bilinear' not in params else params.bilinear
self.batcher = batcher
self.prepare = prepare if prepare else lambda x, y: None
# sanity check
# assert params.classifier in ['LogReg', 'MLP']
# if params.classifier == 'MLP':
# assert params.nhid > 0, 'When using an MLP, \
# you need to set params.nhid>0'
# if not params.usepytorch and params.classifier == 'MLP':
# assert False, 'No MLP implemented in scikit-learn'
self.list_tasks = [
'CR', 'MR', 'MPQA', 'SUBJ', 'SST2', 'SST5', 'TREC', 'MRPC',
'SICKRelatedness', 'SICKEntailment', 'STSBenchmark', 'SNLI',
'ImageCaptionRetrieval', 'STS12', 'STS13', 'STS14', 'STS15',
'STS16', 'DIS', 'PDTB', 'DAT', 'PDTB_EX', 'PDTB_IMEX', 'ABSA_CH',
'ABSA_SP', 'STS_SP'
]
for norm in [True]:
for sim_name in [
'von_mises_correction_tic', 'von_mises_correction_aic'
]:
print('Similarity: {0}'.format(sim_name))
params_senteval = {
'task_path': PATH_TO_DATA,
'usepytorch': True,
'kfold': 10,
'padding': '.'
}
# Word2Vec Google News does not have a . embedding
if word_vec == 'word2vec_GN':
params_senteval['padding'] = 'the'
sim_params = dotdict({
'similarity': sim_name,
'word_vec': word_vec,
'word_count_path': word_count_path,
'norm': norm
})
params_senteval['sim_params'] = sim_params
se = senteval.engine.SE(params_senteval, batcher, prepare)
result = se.eval(transfer_tasks)
result_dict = {'param': dict(sim_params), 'eval': result}
results.append(result_dict)
with open(args.output_path, 'w') as f:
json.dump(results, f)
#transfer_tasks = ['MR', 'CR', 'MPQA', 'SUBJ', 'SST2', 'SST5', 'TREC', 'MRPC',
# 'SICKEntailment', 'SICKRelatedness', 'STSBenchmark',
# 'Length', 'WordContent', 'Depth', 'TopConstituents',
# 'BigramShift', 'Tense', 'SubjNumber', 'ObjNumber']
from embed import Embedder
from fairseq import options
import sentencepiece as spm
from senteval import utils
os.chdir(CODE_DIR)
parser = options.get_generation_parser(interactive=True)
options.add_embed_args(parser)
args = options.parse_args_and_arch(parser)
params = utils.dotdict(params_senteval)
sp = spm.SentencePieceProcessor()
sp.Load(args.sentencepiece)
embedder = Embedder(args)
params.batch_size = 32
params.sp = sp
params.embedder = embedder
params.encoder = args.eval_encoder
params.lang_emb = args.lang_emb
results = {}
s = ENLengthEval(PATH_TO_MY_DATA)
s.do_prepare(params, prepare)
