def main():
parser = argparse.ArgumentParser()
parser.add_argument('--infile',
type=str,
default='../../dataset/T=2_train.json')
parser.add_argument('--outfile',
type=str,
default="../models/sketchadapt/model.ckpt")
parser.add_argument('--epochs', type=int, default=5)
parser.add_argument('--val_split', type=float)
parser.add_argument('-E', type=int, default=20, help='embedding dimension')
parser.add_argument('--nb_inputs', type=int, default=3)
args = parser.parse_args()
problems = json.loads(open(args.infile).read())
max_token_length = util.get_max_token_len(args.infile)
rows_type, rows_val, y = get_XY(problems, args.nb_inputs, max_token_length)
model = Sketchadapt(args.nb_inputs,
args.E,
max_token_length / 5,
batch_size=64,
lr=5e-4)
model.fit(rows_type,
rows_val,
y,
epochs=args.epochs,
validation_split=args.val_split)
print('saving model to ', args.outfile)
model.save(args.outfile)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--infile',
type=str,
default='../../dataset/T=2_train.json')
parser.add_argument('--outfile',
type=str,
default="../models/rubustfill/model.ckpt")
parser.add_argument('--epochs', type=int, default=30)
parser.add_argument('--val_split', type=float)
parser.add_argument('-E', type=int, default=8, help='embedding dimension')
parser.add_argument('--nb_inputs', type=int, default=3)
args = parser.parse_args()
problems = json.loads(open(args.infile).read())
max_token_length = util.get_max_token_len(args.infile)
rows_type, rows_val, y = get_XY(problems, args.nb_inputs, max_token_length)
model = Rubustfill(args.nb_inputs,
args.E,
max_token_length,
batch_size=64,
K=128,
attention=None,
lr=5e-4)
for i in range(args.epochs // 10):
model.fit(rows_type,
rows_val,
y,
epochs=10,
validation_split=args.val_split)
print('saving model to ', args.outfile)
model.save(args.outfile)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--problemfile', type=str, default='../../dataset/T=2_test.json')
parser.add_argument('--predictor', type=str, default='True')
parser.add_argument('--outfile', type=str)
parser.add_argument('--T', type=int, default=2)
parser.add_argument('--mode', type=str,
choices=['dfs', 'sort-and-add'],
default='dfs')
parser.add_argument('--gas', type=int, default=1500)
parser.add_argument('-E', type=int, default=20, help='embedding dimension')
parser.add_argument('--nb_inputs', type=int, default=3)
args = parser.parse_args()
problems = json.loads(open(args.problemfile).read())
# annotate problems with predictions
max_token_length = util.get_max_token_len(args.problemfile)
predictor = sketchadapt.Sketchadapt(args.nb_inputs, args.E, max_token_length/5)
predictor.load()
rows_type, rows_val, y = sketchadapt.get_XY(problems, args.nb_inputs, max_token_length)
sketch_pred = predictor.predict_sketch(rows_type, rows_val)
nb_beam = int(args.gas**0.4)
nb_beam = 50
fs = search.beam_search_sketcher(sketch_pred, int(max_token_length/5), nb_beam) # [batch, nb_beam, T]
print(np.array(fs).shape)
arg_pred_list = [] # [nb_beam, batch, 27]
for i in range(nb_beam):
arg_pred_list.append(predictor.predict_args(rows_type, rows_val, np.array(fs)[:, i, :]))
# print(np.array(fs).shape)
print(np.array(arg_pred_list).shape)
print(np.stack(arg_pred_list, axis=1).shape)
predictions = list(zip(fs[i], np.stack(arg_pred_list, axis=1)[i]) for i in range(len(fs)))
for problem, pred in zip(problems, predictions):
problem['prediction'] = pred
rows = solve_problems(problems, args.T, args.mode, args.gas, nb_beam)
df = pd.DataFrame(rows)
nb_solved = len(df) - sum(df.solution.isnull())
print('summary:')
print('solved {}/{} ({}%)'.format(nb_solved, len(df), nb_solved * 100. / len(df)))
print(df.describe())
if args.outfile:
print('saving results to', args.outfile)
df.to_hdf(args.outfile, 'data')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--problemfile',
type=str,
default='../../dataset/T=2_test.json')
parser.add_argument('--predictor', type=str, default='True')
parser.add_argument('--outfile', type=str)
parser.add_argument('--T', type=int, default=2)
parser.add_argument('--mode',
type=str,
choices=['dfs', 'sort-and-add', 'beam'],
default='beam')
parser.add_argument('--gas', type=int, default=500)
parser.add_argument('-E', type=int, default=8, help='embedding dimension')
parser.add_argument('--nb_inputs', type=int, default=3)
args = parser.parse_args()
problems = json.loads(open(args.problemfile).read())
if args.predictor:
# annotate problems with predictions
predictor = deepcoder_tf.Deepcoder(args.nb_inputs, args.E, K=128)
predictor.load()
rows_type, rows_val, y = deepcoder_tf.get_XY(problems, args.nb_inputs)
predictions = predictor.predict(rows_type, rows_val)
if args.mode == 'beam':
max_token_length = util.get_max_token_len(args.problemfile)
for problem, pred in zip(problems, predictions):
# problem['prediction'] = np.array([np.concatenate([pred * 17/21, np.ones([8, ],dtype=np.float32) * 4.0 / 21], axis=-1) for _ in range(max_token_length)])
# print(problem['prediction'])
rand = np.random.rand(10, 42)
problem['prediction'] = np.array(
[rand[_] / np.sum(rand[_]) for _ in range(10)])
else:
for problem, pred in zip(problems, predictions):
problem['prediction'] = pred
rows = solve_problems(problems, args.T, args.mode, args.gas)
df = pd.DataFrame(rows)
nb_solved = len(df) - sum(df.solution.isnull())
print('summary:')
print('solved {}/{} ({}%)'.format(nb_solved, len(df),
nb_solved * 100. / len(df)))
print(df.describe())
if args.outfile:
print('saving results to', args.outfile)
df.to_hdf(args.outfile, 'data')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--problemfile',
type=str,
default='../../dataset/T=2_test.json')
parser.add_argument('--predictor', type=str, default='True')
parser.add_argument('--outfile', type=str)
parser.add_argument('--T', type=int, default=2)
parser.add_argument('--mode',
type=str,
choices=['dfs', 'sort-and-add', 'beam'],
default='beam')
parser.add_argument('--gas', type=int, default=500)
parser.add_argument('-E', type=int, default=62, help='embedding dimension')
parser.add_argument('--nb_inputs', type=int, default=3)
args = parser.parse_args()
problems = json.loads(open(args.problemfile).read())
if args.predictor:
# annotate problems with predictions
max_token_length = util.get_max_token_len(args.problemfile)
predictor = transformer_rf.Transfill(args.nb_inputs,
args.E,
max_token_length,
K=64)
predictor.load()
rows_type, rows_val, y = transformer_rf.get_XY(problems,
args.nb_inputs,
max_token_length)
predictions = predictor.predict(
rows_type, rows_val,
np.ones_like(y) * (len(impl.ACT_SPACE) - 1))
for problem, pred in zip(problems, predictions):
problem['prediction'] = pred
# print(pred)
rows = solve_problems(problems, args.T, args.mode, args.gas)
df = pd.DataFrame(rows)
nb_solved = len(df) - sum(df.solution.isnull())
print('summary:')
print('solved {}/{} ({}%)'.format(nb_solved, len(df),
nb_solved * 100. / len(df)))
print(df.describe())
if args.outfile:
print('saving results to', args.outfile)
df.to_hdf(args.outfile, 'data')