def main():
args = get_arguments()
print("initializing device ... ", end="", file=sys.stderr, flush=True)
dev = D.Naive() if args.gpu < 0 else D.CUDA(args.gpu)
Device.set_default(dev)
print("done.", file=sys.stderr)
mode = args.mode
prefix = args.model
if mode == "train":
encdec = EncoderDecoder(args.dropout)
encdec.init(args.src_vocab, args.trg_vocab, args.embed, args.hidden)
optimizer = O.Adam()
optimizer.set_weight_decay(1e-6)
optimizer.set_gradient_clipping(5)
train(encdec, optimizer, args, 1e10)
elif mode == "resume":
print("loading model/optimizer ... ",
end="",
file=sys.stderr,
flush=True)
encdec = EncoderDecoder(args.dropout)
encdec.load(prefix + ".model")
optimizer = O.Adam()
optimizer.load(prefix + ".optimizer")
valid_ppl = load_ppl(prefix + ".valid_ppl")
print("done.", file=sys.stderr)
train(encdec, optimizer, args, valid_ppl)
else:
print("loading model ... ", end="", file=sys.stderr, flush=True)
encdec = EncoderDecoder(args.dropout)
encdec.load(prefix + ".model")
print("done.", file=sys.stderr)
test(encdec, args)
def main():
parser = ArgumentParser()
parser.add_argument("mode", help="(train|resume|test)")
parser.add_argument("model_prefix", help="prefix of the model files.")
args = parser.parse_args()
mode = args.mode
prefix = args.model_prefix
print("mode:", mode, file=sys.stderr)
print("prefix:", prefix, file=sys.stderr)
if mode not in ("train", "resume", "test"):
print("unknown mode:", mode, file=sys.stderr)
return
print("initializing device ... ", end="", file=sys.stderr)
sys.stderr.flush()
dev = D.CUDA(0)
Device.set_default(dev)
print("done.", file=sys.stderr)
if mode == "train":
encdec = AttentionalEncoderDecoder()
encdec.init(SRC_VOCAB_SIZE, TRG_VOCAB_SIZE, NUM_EMBED_UNITS,
NUM_HIDDEN_UNITS)
optimizer = O.Adam()
optimizer.set_weight_decay(1e-6)
optimizer.set_gradient_clipping(5)
train(encdec, optimizer, prefix, 1e10)
elif mode == "resume":
print("loading model/optimizer ... ", end="", file=sys.stderr)
sys.stderr.flush()
encdec = AttentionalEncoderDecoder()
encdec.load(prefix + ".model")
optimizer = O.Adam()
optimizer.load(prefix + ".optimizer")
valid_ppl = load_ppl(prefix + ".valid_ppl")
print("done.", file=sys.stderr)
train(encdec, optimizer, prefix, valid_ppl)
else:
print("loading model ... ", end="", file=sys.stderr)
sys.stderr.flush()
encdec = AttentionalEncoderDecoder()
encdec.load(prefix + ".model")
print("done.", file=sys.stderr)
test(encdec)
def main():
parser = ArgumentParser()
parser.add_argument("mode")
parser.add_argument("model_prefix")
args = parser.parse_args()
mode = args.mode
prefix = args.model_prefix
print("mode:", mode, file=sys.stderr)
print("prefix:", prefix, file=sys.stderr)
if mode not in ("train", "resume", "test"):
print("unknown mode:", mode, file=sys.stderr)
return
print("initializing device ... ", end="", file=sys.stderr)
sys.stderr.flush()
dev = D.Naive() # = D.CUDA(0)
Device.set_default(dev)
print("done.", file=sys.stderr)
if mode == "train":
encdec = EncoderDecoder("encdec", SRC_VOCAB_SIZE, TRG_VOCAB_SIZE,
NUM_EMBED_UNITS, NUM_HIDDEN_UNITS,
DROPOUT_RATE)
trainer = T.Adam()
trainer.set_weight_decay(1e-6)
trainer.set_gradient_clipping(5)
train(encdec, trainer, prefix, 1e10)
elif mode == "resume":
print("loading model/trainer ... ", end="", file=sys.stderr)
sys.stderr.flush()
encdec = EncoderDecoder.load("encdec", prefix + '.')
trainer = T.Adam()
trainer.load(prefix + ".trainer.config")
valid_ppl = load_ppl(prefix + ".valid_ppl.config")
print("done.", file=sys.stderr)
train(encdec, trainer, prefix, valid_ppl)
else: # mode == "test"
print("loading model ... ", end="", file=sys.stderr)
sys.stderr.flush()
encdec = EncoderDecoder.load("encdec", prefix + '.')
print("done.", file=sys.stderr)
test(encdec)