from lib.ProcessEmbeddings import WordEmbeddings
from tools.Blogger import Blogger
import os
logger = Blogger()
CLASSIFICATION_TASKS = ["MR", "CR", "SUBJ", "MPQA", "SST5", "TREC", "MRPC"]
SIMILARITY_TASKS = [
"SICKRelatedness", "STS12", "STS13", "STS14", "STS15", "STS16"
]
if __name__ == "__main__":
WE = WordEmbeddings(vector_file="embeds/glove.6B.300d.txt")
WE.sparsify("ckpt/glove3000/ckpt-8000")
WE.subract_mean()
logger.status_update("Running SentEval tasks...")
WE.SentEval(
tasks=CLASSIFICATION_TASKS,
save_summary=True,
summary_file_name="glove_wta_3000.json",
)
type=str2bool,
default=False,
help="Whether to use pytorch as classifier",
)
parser.add_argument("-batch_size",
type=int,
default=128,
help="Batch size for classification")
parser.add_argument("-epoch_size", type=int, default=2, help="Epoch size")
values = parser.parse_args()
logger.green(values)
PATH_TO_VEC = values.path
# Set params for SentEval
params_senteval = {
"task_path": PATH_TO_DATA,
"usepytorch": values.pytorch,
"kfold": 5,
}
params_senteval["classifier"] = {
"nhid": 0,
"optim": "rmsprop",
"batch_size": values.batch_size,
"tenacity": 3,
"epoch_size": values.epoch_size,
}
se = senteval.engine.SE(params_senteval, batcher, prepare)
result = se.eval(values.tests)
for k in result:
logger.status_update("{}: {}".format(k, result[k]["acc"]))
print()