x_train = scaler.transform(x_train)
x_test = scaler.transform(x_test)
y = y * 2 - 1
print("x_train", x_train.shape, x_train.min(), x_train.max())
print("x_test", x_test.shape, x_test.min(), x_test.max())
print("y", y.shape, y.min(), y.max())
x_test = np.clip(x_test, a_min=-1, a_max=1)
print("x_test", x_test.shape, x_test.min(), x_test.max())
mlp = MLP(layer_size=[x_train.shape[1], 28, 28, 28, 1],
regularization=1,
output_shrink=0.1,
output_range=[-1, 1],
loss_type="hardmse")
"""
train 15 epoches, batch_size=1, SGD
"""
mlp.train(x_train,
y,
verbose=2,
iteration_log=20000,
rate_init=0.08,
rate_decay=0.8,
epoch_train=15,
epoch_decay=1)
pred = mlp.predict(x_test)
pred = pred.reshape(-1)
scaler = MinMaxScaler(feature_range=(0, 1)).fit(
np.concatenate((x_train_data, x_test_data), axis=0))
x_train_data = scaler.transform(x_train_data)
x_test_data = scaler.transform(x_test_data)
# In[ ]:
import ultimate
from ultimate.mlp import MLP
epoch_train = 1000
mlp = MLP(layer_size=[x_train_data.shape[1], 28, 28, 1],
regularization=1,
output_shrink=0.1,
output_range=[0, 1],
loss_type="hardmse")
mlp.train(x_train_data,
p_train_data,
iteration_log=20000,
rate_init=0.08,
rate_decay=0.8,
epoch_train=epoch_train,
epoch_decay=1)
# In[ ]:
predictions = np.round(mlp.predict(x_test_data).reshape(-1))
# In[ ]:
#X_train = np.load('X_train_hok_msm.npy')
#y = np.load('y_hok_msm.npy')
epoch_decay = 2
epoch_train = epoch_decay * 18
rate_init = 0.008
hidden_size = 32
verbose = 1
activation = 'a2m2l'
leaky = -0.001
mlp = MLP(
layer_size=[X_train.shape[1], hidden_size, hidden_size, hidden_size, 1],
activation=activation,
leaky=leaky,
bias_rate=[],
regularization=1,
importance_mul=0.0001,
output_shrink=0.1,
output_range=[-1, 1],
loss_type="hardmse")
mlp.train(X_train,
y,
verbose=verbose,
importance_out=True,
iteration_log=20000,
rate_init=rate_init,
rate_decay=0.8,
epoch_train=epoch_train,
epoch_decay=epoch_decay)
X_test, _ = feature_engineering(False)
gc.collect()
X_train['totalDistance'] = X_train['walkDistance'] + X_train[
'rideDistance'] + X_train['swimDistance']
X_test['totalDistance'] = X_test['walkDistance'] + X_test[
'rideDistance'] + X_test['swimDistance']
del X_train['walkDistance'], X_train['rideDistance'], X_train[
'swimDistance']
del X_test['walkDistance'], X_test['rideDistance'], X_test['swimDistance']
gc.collect()
scaler = preprocessing.MinMaxScaler(copy=False).fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(X_test)
mlp_param = {
'loss_type': 'mse',
'layer_size': [X_train.shape[1], 8, 8, 8, 1],
'output_range': [0, 1],
'importance_out': True,
'rate_init': 0.02,
'epoch_train': 5,
'epoch_decay': 10,
'verbose': 1,
}
mlp = MLP(mlp_param)
mlp.fit(X_train, y_train)
print("feature importances:", mlp.feature_importances_)
pred = mlp.predict(X_test)
save_csv(pred, test)
end_time = time.time()
print('The total of time is', end_time - start_time)