if __name__ == "__main__":
parser = argparse.ArgumentParser(
description=
"Benchmark CNN on a bipartite ranking task (answer selection)")
parser.add_argument("-N", help="GloVe dim", type=int, default=300)
parser.add_argument("--wang",
help="whether to run on Wang inst. of YodaQA dataset",
type=int,
default=0)
parser.add_argument("--params",
help="additional training parameters",
type=str,
default='')
args = parser.parse_args()
glove = emb.GloVe(N=args.N)
if args.wang == 1:
s0, s1, y, vocab, gr = load_set('data/anssel/wang/train-all.csv',
glove)
s0t, s1t, yt, _, grt = load_set('data/anssel/wang/dev.csv', glove,
vocab)
else:
s0, s1, y, vocab, gr = load_set(
'data/anssel/yodaqa/curatedv1-training.csv', glove)
s0t, s1t, yt, _, grt = load_set('data/anssel/yodaqa/curatedv1-val.csv',
glove, vocab)
kwargs = eval('dict(' + args.params + ')')
model = prep_model(glove, vocab, oact='linear', **kwargs)
model.compile(
loss={'score': ranknet},
ev.eval_hypev(prediction_t, s0t, grt['score'], 'Val')
dump_questions(s0, s1, gr['score'], prediction, 'Train')
dump_questions(s0t, s1t, grt['score'], prediction_t, 'Val')
return model
if __name__ == "__main__":
modelname, trainf, valf = sys.argv[1:4]
# modelname, trainf, valf = 'avg', 'data/hypev/argus/argus_train.csv', 'data/hypev/argus/argus_test.csv'
params = sys.argv[4:]
module = importlib.import_module('.' + modelname, 'models')
conf, ps, h = config(module.config, params)
runid = '%s-%x' % (modelname, h)
print('RunID: %s (%s)' % (runid, ps))
if conf['embdim'] is not None:
print('GloVe')
glove = emb.GloVe(N=conf['embdim'])
else:
glove = None
print('Dataset')
s0, s1, y, vocab, gr = load_set(trainf)
s0t, s1t, yt, _, grt = load_set(valf, vocab)
pickle.dump(vocab, open('vocab.txt', 'wb'))
train_and_eval(runid, module.prep_model, conf, glove, vocab, gr, s0, grt,
s0t)
params = sys.argv[6:]
load_s1texts(s1f)
model_module = importlib.import_module('.' + modelname, 'models')
task_module = importlib.import_module('.' + taskname, 'tasks')
task = task_module.task()
conf, ps, h = config(model_module.config, task.config, params)
task.set_conf(conf)
print(ps)
# TODO we should be able to get away with actually *not* loading
# this at all!
if conf['embdim'] is not None:
print('GloVe')
task.emb = emb.GloVe(N=conf['embdim'])
else:
task.emb = None
print('Dataset')
task.load_vocab(vocabf)
print('Model')
task.c['skip_oneclass'] = False # load_samples() returns oneclass
model = task.build_model(model_module.prep_model)
print(weightsf)
model.load_weights(weightsf)
print("Running...")
app.run(port=conf.get('port', 5051),
args = sys.argv[1:]
if args[0] == '--revocab':
revocab = True
args = args[1:]
else:
revocab = False
trainf, valf, testf, dumptrainf, dumpvalf, dumptestf, vocabf = args
if revocab:
vocab = Vocabulary(sentence_gen([trainf]), count_thres=2)
print('%d words' % (len(vocab.word_idx)))
pickle.dump(vocab, open(vocabf, "wb"))
else:
vocab = pickle.load(open(vocabf, "rb"))
print('%d words' % (len(vocab.word_idx)))
glove = emb.GloVe(N=300) # XXX: hardcoded
print('Preprocessing train file')
si0, si1, sj0, sj1, f0_, f1_, labels = load_set(trainf, vocab, glove)
pickle.dump((si0, si1, sj0, sj1, f0_, f1_, labels), open(dumptrainf, "wb"))
print('Preprocessing validation file')
si0, si1, sj0, sj1, f0_, f1_, labels = load_set(valf, vocab, glove)
pickle.dump((si0, si1, sj0, sj1, f0_, f1_, labels), open(dumpvalf, "wb"))
print('Preprocessing test file')
si0, si1, sj0, sj1, f0_, f1_, labels = load_set(testf, vocab, glove)
pickle.dump((si0, si1, sj0, sj1, f0_, f1_, labels), open(dumptestf, "wb"))
import pysts.kerasts.blocks as B
from pysts.kerasts.callbacks import AnsSelCB
from pysts.kerasts.objectives import ranknet
from pysts.hyperparam import RandomSearch
import anssel_train
if __name__ == "__main__":
modelname, trainf, valf = sys.argv[1:4]
module = importlib.import_module('.' + modelname, 'models')
s0, s1, y, vocab, gr = anssel_train.load_set(trainf)
s0t, s1t, yt, _, grt = anssel_train.load_set(valf, vocab)
glove = emb.GloVe(300) # XXX hardcoded N
rs = RandomSearch(modelname + '_rlog.txt',
dropout=[1 / 2, 2 / 3, 3 / 4],
inp_e_dropout=[1 / 2, 3 / 4, 4 / 5],
l2reg=[1e-4, 1e-3, 1e-2],
cnnact=['tanh', 'tanh', 'relu'],
cnninit=['glorot_uniform', 'glorot_uniform', 'normal'],
cdim={
1: [0, 0, 1 / 2, 1, 2],
2: [0, 0, 1 / 2, 1, 2, 0],
3: [0, 0, 1 / 2, 1, 2, 0],
4: [0, 0, 1 / 2, 1, 2, 0],
5: [0, 0, 1 / 2, 1, 2]
},
project=[True, True, False],
class ElasticScore(Feature):
def __init__(self, answer, i):
Feature.set_type(self, rel)
Feature.set_name(self, 'Elastic score')
Feature.set_value(self, answer.sources[i].elastic)
import pickle
import importlib
import pysts.embedding as emb
from argus.keras_build import config, build_model, load_sent
module = importlib.import_module('.' + 'rnn', 'models')
conf, ps, h = config(module.config, [])
print 'loading sts model, glove'
glove = emb.GloVe(N=50)
vocab = pickle.load(open('sources/vocab.txt'))
print 'glove loaded'
model = build_model(glove, vocab, module.prep_model, conf)
model.load_weights('sources/models/keras_model.h5')
print 'sts model loaded'
class STS_NN(Feature):
"""
Keras models from brmson/dataset-sts
"""
def __init__(self, answer, i):
Feature.set_type(self, clas + rel)
Feature.set_name(self, 'STS_NN')
gr = load_sent(answer.q.text, answer.sources[i].sentence, vocab)
