parser.add_argument("--lr", type=float, default=2e-3,
help="Initial learning rate.")
parser.add_argument("--lr_decay_by", type=float, default=0.75,
help="The ratio to multiply the learning rate by every "
"time the learning rate is decayed.")
parser.add_argument("--lr_decay_every", type=float, default=1,
help="Decay the learning rate every _ epoch(s).")
parser.add_argument("--dev_every", type=int, default=1000,
help="Run evaluation on the dev split every _ training "
"batches.")
parser.add_argument("--save_every", type=int, default=1000,
help="Save checkpoint every _ training batches.")
parser.add_argument("--embed_dropout", type=float, default=0.08,
help="Word embedding dropout rate.")
parser.add_argument("--mlp_dropout", type=float, default=0.07,
help="SNLIClassifier multi-layer perceptron dropout "
"rate.")
parser.add_argument("--no-projection", action="store_false",
dest="projection",
help="Whether word embedding vectors are projected to "
"another set of vectors (see d_proj).")
parser.add_argument("--predict_transitions", action="store_true",
dest="predict",
help="Whether the Tracker will perform prediction.")
parser.add_argument("--force_cpu", action="store_true", dest="force_cpu",
help="Force use CPU-only regardless of whether a GPU is "
"available.")
FLAGS, unparsed = parser.parse_known_args()
tfe.run(main=main, argv=[sys.argv[0]] + unparsed)
help="Word embedding dropout rate.")
parser.add_argument(
"--mlp_dropout",
type=float,
default=0.5,
help="ChemprotClassifier multi-layer perceptron dropout "
"rate.")
parser.add_argument("--no-projection",
action="store_false",
dest="projection",
help="Whether word embedding vectors are projected to "
"another set of vectors (see d_proj).")
parser.add_argument("--predict_transitions",
action="store_true",
dest="predict",
help="Whether the Tracker will perform prediction.")
parser.add_argument(
"--force_cpu",
action="store_true",
dest="force_cpu",
help="Force use CPU-only regardless of whether a GPU is "
"available.")
parser.add_argument("--test_bool",
action="store_true",
dest="test_bool",
help="For test")
FLAGS, unparsed = parser.parse_known_args()
tfe.run(main=main, argv=["--data_root chemprot-data --logdir tmpLog"])
with Measure('apply', times):
optimizer.apply_gradients(
zip(grad, cvae.variables),
global_step=tf.train.get_or_create_global_step())
# Printing output
print("#ALL")
tot_avg, tot_sum = Measure.print_times(times)
print("\nNodes: {0:.1f} ({1:.1f})".format(np.mean(node_count),
np.std(node_count)))
print("Depths: {0:.1f} ({1:.1f})".format(np.mean(node_depth),
np.std(node_depth)))
print("Arities: {0:.1f} ({1:.1f})".format(np.mean(node_arity),
np.std(node_arity)))
print((np.sum(node_count) / tot_sum),
FLAGS.batch_size * FLAGS.benchmark_runs / tot_sum)
# print("\n#CVAE")
# Measure.print_times(cvae.loss_times)
#
# print("\n#DEC")
# Measure.print_times(cvae._det_decoder.__class__.times)
if __name__ == "__main__":
import warnings
warnings.filterwarnings(
"ignore") # deprecated stuff in TF spams the console
define_flags()
tfe.run()
parser.add_argument("--lr", type=float, default=2e-3,
help="Initial learning rate.")
parser.add_argument("--lr_decay_by", type=float, default=0.75,
help="The ratio to multiply the learning rate by every "
"time the learning rate is decayed.")
parser.add_argument("--lr_decay_every", type=float, default=1,
help="Decay the learning rate every _ epoch(s).")
parser.add_argument("--dev_every", type=int, default=1000,
help="Run evaluation on the dev split every _ training "
"batches.")
parser.add_argument("--save_every", type=int, default=1000,
help="Save checkpoint every _ training batches.")
parser.add_argument("--embed_dropout", type=float, default=0.08,
help="Word embedding dropout rate.")
parser.add_argument("--mlp_dropout", type=float, default=0.07,
help="SNLIClassifier multi-layer perceptron dropout "
"rate.")
parser.add_argument("--no-projection", action="store_false",
dest="projection",
help="Whether word embedding vectors are projected to "
"another set of vectors (see d_proj).")
parser.add_argument("--predict_transitions", action="store_true",
dest="predict",
help="Whether the Tracker will perform prediction.")
parser.add_argument("--force_cpu", action="store_true", dest="force_cpu",
help="Force use CPU-only regardless of whether a GPU is "
"available.")
FLAGS, unparsed = parser.parse_known_args()
tfe.run(main=main, argv=[sys.argv[0]] + unparsed)