def regression():
inp_dim, out_dim = 10, 5
params = {"max_depth": args.depth, "lr": args.lr, 'loss': b"mse"}
booster = GBDTMulti(LIB, out_dim=out_dim, params=params)
x_train, y_train = np.random.rand(10000, inp_dim), np.random.rand(10000, out_dim)
x_valid, y_valid = np.random.rand(10000, inp_dim), np.random.rand(10000, out_dim)
booster.set_data((x_train, y_train), (x_valid, y_valid))
booster.train(20)
booster.dump(b"regression.txt")
def classification():
inp_dim, out_dim = 10, 5
params = {"max_depth": args.depth, "lr": args.lr, 'loss': b"ce"}
booster = GBDTMulti(LIB, out_dim=out_dim, params=params)
x_train = np.random.rand(10000, inp_dim)
y_train = np.random.randint(0, out_dim, size=(10000, )).astype("int32")
x_valid = np.random.rand(10000, inp_dim)
y_valid = np.random.randint(0, out_dim, size=(10000, )).astype("int32")
booster.set_data((x_train, y_train), (x_valid, y_valid))
booster.train(20)
booster.dump(b"classification.txt")
def classification(data, meta, depth, lr, k, one_side):
print("depth: {}, lr: {}, k: {}, one_side: {}".format(
depth, lr, k, one_side))
p = {
'max_depth': depth,
'max_leaves': int(0.75 * 2**depth),
'topk': k,
'loss': b"ce",
'gamma': 1e-3,
'num_threads': 8,
'max_bins': meta['bin'],
'lr': lr,
'reg_l2': 1.0,
'early_stop': 25,
'one_side': one_side,
'verbose': False,
'min_samples': 16
}
m = GBDTMulti(LIB, out_dim=meta['out'], params=p)
x_train, y_train, x_test, y_test = data
m.set_data((x_train, y_train), (x_test, y_test))
t = time.time()
m.train(ROUND)
t = time.time() - t
if args.time == 1:
print("Average time: {:.3f}".format(t / ROUND))
else:
print("Total time: {:.3f}".format(t))
del m
def regression_multi(data, meta, depth, lr):
p = {
'max_depth': depth,
'max_leaves': int(0.75 * 2**depth),
'topk': 0,
'loss': b"mse",
'gamma': 1e-6,
'num_threads': args.workers,
'max_bins': meta['bin'],
'lr': lr,
'reg_l2': 1.0,
'early_stop': 0,
'one_side': True,
'verbose': False,
'hist_cache': 48,
'min_samples': 4
}
m = GBDTMulti(LIB, out_dim=meta['out'], params=p)
x_train, y_train, x_test, y_test = data
m.set_data((x_train, y_train))
t = time.time()
m.train(ROUND)
t = time.time() - t
del m
return t
def train_gbdt_multi(data, meta):
depth = cfg.Depth[args.mode][args.data]
lr = cfg.Learning_rate[args.mode][args.data]
p = {'max_depth': depth, 'max_leaves': int(0.75 * 2 ** depth), 'topk': 0, 'loss': b"mse",
'gamma': GAMMA, 'num_threads': 8, 'max_bins': meta['bin'], 'lr': lr, 'reg_l2': 1.0,
'early_stop': 25, 'one_side': True, 'verbose': False, 'hist_cache': 48,
'min_samples': 8}
m = GBDTMulti(LIB, out_dim=meta['out'], params=p)
x_train, y_train, x_test, y_test = data
m.set_data((x_train, y_train), (x_test, y_test))
m.train(ROUND)
preds = m.predict(x_test)
del m
np.save("result/gbdtm", preds)
def train_gbdt_multi(data, meta):
depth = cfg.Depth[args.mode][args.data]
lr = cfg.Learning_rate[args.mode][args.data]
p = {'max_depth': depth, 'max_leaves': int(0.75 * 2 ** depth), 'topk': args.k, 'loss': b"ce",
'gamma': GAMMA, 'num_threads': num_threads, 'max_bins': meta['bin'], 'lr': lr, 'reg_l2': 1.0,
'early_stop': 25, 'one_side': True, 'verbose': True,
'min_samples': min_samples}
m = GBDTMulti(LIB, out_dim=meta['out'], params=p)
x_train, y_train, x_test, y_test = data
m.set_data((x_train, y_train), (x_test, y_test))
m.train(ROUND)